KNOWLEDGE TO LEAD by Education Association for Software Empowerment

It has power to create common good and value in society. As a youth leadership organization, HOBY Korea is committed to encouraging young leaders to explore their intellectual curiosity and share their knowledge to benefit others. ‘KNOWLEDGE TO LEAD’ is one of the many platforms HOBY Korea runs to fulfill this commitment. As an interdisciplinary journal for high school students, it provids high school students with an opportunity to publish manuscripts in various research fields to promote intellectual curiosity, broaden the perspective towards social phenomena and global issues, and enhance creative problem-solving skills. The manuscripts published in Volume 1 Issue 1 were the outcome of HOBY Korea’ ACADEMIC RESEARCH SEMINAR FOR GLOBAL YOUTH 2020. Though our first step is small, we will develop this journal as one of the most prominent youth journals by setting the bar higher with each issue.

TABLE OF CONTENTS ARTICLES – LEGAL STUDIES/EAST ASIAN STUDIES

SOUTH KOREA’S SURVIVAL ON A TIGHTROPE BETWEEN THE US AND CHINA: THE IMPACTS OF THE CLASH BETWEEN THE US AND CHINA ON SOUTH KOREA, IN COMPARISON TO JAPAN – EDWARD KIM 2 MILITARY COOPERATION WITH SOUTH KOREA AND JAPAN: WHAT THE UNITED STATES WANTS, BUT CANNOT EASILY ACHIEVE - IN YEOL YEO

ARTICLES – MEDICINE

IMAGE PROCESSING OF BRAIN CT IN ACUTE CEREBRAL INFARCTION – BEOMSEOK KIM

A COMPARISON OF GAUSSIAN FILTERING AND FOURIER TRANSFORM TECHNIQUES FOR IMAGE DENOISING – KYUSEUNG SHIN

EXTRACTING RIB STRUCTURE FROM TWO-DIMENSIONAL CHEST X-RAY USING HIGH DYNAMIC RANGE IMAGING METHOD IN OPENCV – JIMIN HAN, SUMIN HAN

INVERSE PROBLEMS AND NOISE REDUCTION IN IMAGING BY POST-IMAGE PROCESSING – SUMIN HAN

A STUDY ON THE COMPOSITION OF CERTAIN ANTISEPTIC AND ANTIBACTERIAL BACTERIA AND GINSENG FERMENTED GINSENG SOLUTION – JI SOO HWANG HWANG

South Korea’s Survival on a Tightrope between the US and China: The Impacts of the Clash between the US and China on South Korea, in comparison to Japan Edward Kim Harrow School- UK edjisookim@gmail.com THESIS Northeast Asia is a both a pivotal and a highly volatile region in the global political and economic landscape. The balance of power in the region rested upon the United States’ political and economic influence over its allies, which include South Korea and Japan. The emergence of the US as the centre of the democratic capitalist world firmly established its influence in Northeast Asia, emerging from the turbulence of the conflict-torn mid-twentieth century. The political and military influence of the US became more visible in the 21st century as the Cold War ended at the end of the twentieth century. In addition to the power coming from high politics, the US had also established a firm economic stronghold in Northeast Asia, which was exemplified by the trade statistics of South Korea and Japan in the year 1990. The US was the largest export partner for South Korea with 29.91% of the global export partner share and the secondlargest importer (only second to Japan) with 24.26% of the import partner share. Similarly, the US was both the largest import partner of Japan with 22.59% of the import partner share and the largest export partner with 31.69% of the export partner share. In 1990, the balance of power rested on the US with regards to both the high politics and low politics. However, China’s rapid economic growth following the diplomatic move away from its isolationist policies pose a direct challenge to the established power of the US. China has already assumed regional economic dominance. The status of the primary trading partner affords china a political advantage wherein it controls both a nation’s supply chain and its export sources, especially for the South Korean and Japanese economies, which rely greatly on exports. The economic expansion is echoed by China’s growing presence of high politics in Northeast Asia, which has challenged the established American influence in South Korea. The clash between the US and China is evolving into a power struggle for Northeast Asia, and I plan to investigate the effects of this China-US clash in the high and low politics of South Korea, comparing them to those of Japan when appropriate. In doing so, I will attempt to underscore the intrinsic connections between history, military, economics, and politics.

HISTORICAL BACKGROUND To understand the complications of the relationship between China and the United States, and its impact on South Korea (and Japan), the historical background must be acknowledged. The American influence on South Korea and Japan began after the allied victory in the Second World War. The vacuum of power created by the fall of the Japanese Empire in 1945 was largely filled by the victors of the war. The United States and the Soviet Union each formed a polarising sphere of influence. This bipolar division of the political systems during the Cold War gave rise to American influence in South Korea and Japan. China, afterall, had been politically and economically disconnected from the rest of the world due to its isolationist policy and hostility towards the capitalist economies. However, the Chinese political stance towards engagement with the US changed when the U.S. national security adviser, Henry Kissinger, secretly visited China in 1971. China's ‘Reform and opening up’ program came into fruition so that after its launch in 1978, China’s economic potential began to be realized. The consequences of the Chinese economic growth and its weight in trade markets shifted the economic landscape of not only Northeast Asia but of the world. Historical examples reflect American politics responding to such economic threats to the US. The US-Japan currency war exemplifies such incidence. The Plaza Accord of 1985 resulted

in a sharp appreciation of the yen, which continued until 19951. This meant that Japan suffered from a trade deficit as Japanese firms lost competitiveness in foreign markets and imports were more attractive to the Japanese consumption market. The stagnation of the Japanese economy should be considered when comparing to the economic rivalry forming between China and the US. The US has been suffering from a chronic bilateral trade deficit with China, drawing parallels to the economic relationship with Japan in the 1980s. Yet, one notable distinction is the lack of political leverage for the US. The US, as stated above, had political and economic means to respond to Japan, but the US does not seem to have powerful leverage toward China. During the same period, the South Korean economy significantly grew to become the twelfth largest economy in the world by GDP in 20192. Export’s share of South Korea’s GDP increased from 25.9% in 1995 to 56.3% in 20123, led by the successful expansion of the shipbuilding, electronics, and automobile sectors. THEORETICAL BACKGROUND OF HIGH AND LOW POLITICS The impact of the US-China clash on South Korea will be assessed regarding high and low politics. The concept of high politics refers to the factors that are directly relevant to the states’ very survival. The factors, ‘or ingredients’, of a state’s survival were organised by Thomas Hobbes, but the term was re-introduced during the Cold War when political and military factors posed a direct possibility of nuclear extinction. Low politics, in contrast to high politics, refers to matters that are not directly essential for the immediate survival of the state, namely the economy and social affairs. In this paper, the impact of the US-China clash on South Korea will be first addressed concerning low politics with a focus on economics, and will then be examined regarding high politics, the international relations regarding the security of the nation. Even though modern international relations are highly interdependent with low politics4, I divided the political situation of South Korea into the two distinctions to compare the trade-offs for each policy for South Korea. However, international relations in Northeast Asia will prove that the distinctions of high and low politics are merely artificial, and the factors are largely interwoven. THE IMPACT OF THE US-CHINA CLASH ON SOUTH KOREAN LOW POLITICS The implications of the US-China clash on the low politics of South Korea can be interpreted through its economic structure. The primary concern for the South Korean economy (and the Japanese economy) must be their vulnerability to Chinese policies or retaliation. China overtook Japan as the largest economy in Asia as a result of sustained high growth rates lasting for over three decades. The average annual real Gross Domestic Product (GDP) growth rate of 9.5% meant that the Chinese economy doubled in size every eight years on average5. This exponential economic growth is reflected by the size of the Chinese economy, which is closing the gap with the American and European economies. China is the second-largest national economy in the world with a GDP of over $14.3 trillion as of 2019. The higher growth rates of the Chinese economy threaten to further close the gap with the American economy and establish its global presence as a trading partner. The significance of the Chinese economy as a trading partner is especially evident in the South Korean and Japanese economies. The significance of the Chinese economy on the South Korean and Japanese economies are manifested in the trade statistics of the respective nations. The Chinese market is the largest export destination of South Korea, worth approximately $160 billion in 2018. This accounted for almost 26% of South Korea’s total exports in the year, while the US made up only 11.9% of the export sales, as shown below.

Graph from Ronald McKinnon and Zhao Liu. “Modern Currency Wars: The United States versus Japan” Asian Development Bank Institute. No. 437.

Data from International Monetary Fund estimates.

3Iwulska, Aleksandra.

“Golden Growth: Restoring the lustre of the European economic model: Country Benchmarks”, World Bank, 2012.

Jackson, Robert H. and Georg Sørensen. 2007. “Introduction to International Relations: Theories and approaches. 3rd ed.”: Oxford University Press, p. 106.

Wayne M. Morrison. “China’s Economic Rise: History, Trends, Challenges, and Implications for the United States” Congressional Research Service, www.everycrsreport.com/files/20190625_RL33534_088c5467dd11365dd4ab5f72133db289fa10030f.pdf.

As the Chinese export market provides more than double the export revenue that the American market provides, the Chinese market is the most valuable export destination for South Korea. It is, therefore, extremely important for South Korea to maintain friendly relations with China for its economic survival. The effects of Chinese economic retaliation with a political agenda on South Korea have been exemplified by the aftermath of the deployment of the THAAD missile defence system in 2016. The introduction of an American missile defence system in the proximity of Chinese airspace sparked tensions and China mounted economic retaliation, which is estimated to have cost 16,200 billion won7. The decreased number of Chinese tourists in South Korea was especially evident as the number of Chinese tourists decreased from 8.07 million in 2016 to 4.17 million in 20178. The THAAD incident served to remind South Korea of the leverage held by the Chinese in view of the loss of Chinese tourists, the removal of Chinese investments from the South Korean financial market, non-tariff barriers to trade, and the boycotting of Korean exports. These came as daunting prospects for the South Korean economy. We can see that South Korea’s reliance on and vulnerability to the Chinese economy is not unique when considering that China was the largest export destination of Japan in 2018 with 19.4% of its export share in 2018, having overtaken the US as the largest trading partner9. The heavy reliance on the Chinese market puts Japan in a similar situation to South Korea whereby the economy is at risk from retaliation by China. The loss of the Chinese export market resulting from the government’s political decisions would be the worst consequence for many South Korean and Japanese firms. On top of China’s leverage power as the largest trading partner, the dominance of Chinese supply chains poses an even graver threat to the security of South Korean and Japanese economies. The COVID-19 pandemic has highlighted the dangers of sourcing a large proportion of the supply chain from a single foreign economy. South Korea and Japan are no exceptions in realising these dangers. As neighbours with fragile political links with China, their reliance on China for imports and supply chains has extreme significance. China is by far the largest import origin for both South Korea and Japan with a respective 21.1%10 and 23.8%11 shares of the total imports for the two nations. The most critical facet of the Chinese supply chains for South Korea could be the rare earth metal imports that high-tech industries require. Rare earth metals and alloys are used in computer memory, rechargeable batteries, and magnets. As critical components of electronic devices, cars, and defence systems, these rare earth metals are essential for the South Korean economy, which relies heavily on machinery for their export revenues.

Data and graph from “South Korea (KOR) Exports, Imports, and Trade Partners.” Observatory of Economic Complexity, oec.world/en/profile/country/kor/. 7“ch'oegŭn chungguk kyŏngjejejae p'agŭp'yogwa ch'ujŏng” [The Ripple Effects of Recent Chinese Economic Sanctions]: KDI Korea Development Institute. (eiec.kdi.re.kr/policy/domesticView.do?ac=0000139311). 8Korea 9

Tourism Organization. http://kto.visitkorea.or.kr/viewer/view.kto?id=66056&type=bd Data from https://oec.world/en/profile/country/jpn.

10 11

“South Korea (KOR) Exports, Imports, and Trade Partners.” Observatory of Economic Complexity, oec.world/en/profile/country/kor/. “Japan (JPN) Exports, Imports, and Trade Partners.” Observatory of Economic Complexity, oec.world/en/profile/country/jpn/.

The overwhelming reliance on high-tech products of the South Korean export market results in a necessity for rare-earth metals supplies from China, without which major elements of the economy could be rendered inactive and futile. Furthermore, it is currently difficult for South Korea to find a quick alternate source of rare earth metal imports. In 2019, China had dominated the global rare earth mining quantity with 132000 tonnes, out of the global total of 210000 tonnes13. The near monopolistic Chinese dominance in global export, with an estimated 62.8% share of the global total, render it difficult, if not impossible, to replace the quantity mined outside China A Chinese embargo of rare-earth metals, albeit an unlikely and extreme scenario, would devastate unbearably large proportions of the economies of South Korea and Japan. Thus, China has an incredibly potent card in low politics for South Korea, as well as for Japan. Yet such feared Chinese economic aggression against South Korea is unlikely due to the mutual benefits in trade, and may only result from a direct threat to China’s security. Some grandiose threats from China were not supported by action, proving to have been only an empty threat for political agenda. South Korea should attempt to minimise the economic consequences from the US-China clash through careful political manoeuvring with dexterity. However, the feasibility of complete avoidance of economic pressure from China appears to be difficult if the political pressures intensify. POTENTIAL BENEFITS FOR SOUTH KOREAN LOW POLITICS However, the second implication of conflict between China and the US is rather beneficial for South Korea (and Japan). The American trend of diversifying supply chains away from China can result in more export opportunities for South Korean and Japanese firms. The rivalry between the economies of South Korea, Japan, and China arise from these countries’ overlapping economic focus on high-tech industries. There are long-term benefits to be reaped by South Korean and Japanese tech companies if Chinese companies lose their competitiveness in the global market through the trade war with the US. Of course, some might argue that this will make China stronger in the long run by rendering it less dependent to externalities, but I believe that the Chinese economy would not be able to enjoy the same degree of prosperity internally without international exports. For example, the market share of 5G telecommunication equipment held by Samsung drastically increased as a result of the banning of the companies’ products by the US and the UK diminished Huawei’s market share. Samsung’s revenue from the growth in the 5g market is predicted to be of great value as it will be allegedly “be responsible for half of the Japanese carrier’s near $4 billion 5G infrastructure across Japan,” while providing the equipment for KDDI14. A further decrease in the competitiveness of Huawei, the leading market shareholder of mobile infrastructure in 2017, will most definitely open doors for export opportunities for other telecommunication companies, including Samsung.

12 13

“South Korea (KOR) Exports, Imports, and Trade Partners.” Observatory of Economic Complexity, (oec.world/en/profile/country/kor/). Data from “MINERAL COMMODITY SUMMARIES 2020” U.S. Geological Survey. (pubs.usgs.gov/periodicals/mcs2020/mcs2020.pdf).

14https://pulsenews.co.kr/view.php?year=2019&no=785215#:~:text=Samsung%20Electronics%20will%20be%20the,billion%205G%20infrastru

cture%20across%20Japan.

The structural rivalry between the Chinese and South Korean economies is not limited to the 5g market. Both South Korea and Japan are often cited as major trade partners in the move away from China. In 2019 April, the US State Department listed South Korea and Japan among countries with whom the US have been communicating for establishing new supply chains. Japan and South Korea are currently the 4th and 6th largest exporters to the US, and the trade volume is likely to increase as the US looks for more ‘secure’ supply chains. This pattern is also reflected by the Indian consumer market, which is turning towards South Korea amidst political and military disputes against China. Thus, South Korea may be able to reap long-term benefits from the US-China trade war by creating more export opportunities despite shortterm difficulties. Even though the high labour costs of South Korea will not provide an appropriate alternative for factory locations, Korean tech exports may enjoy less competition and boosted sales. THE IMPACTS OF THE US-CHINA CLASH ON SOUTH KOREAN HIGH POLITICS The relationship between the US and China will develop into a long-lasting competition. This is the view shared by the South Korean government’s Institute of Foreign Affairs and National Security15. It further establishes that Chinese development and expansion will most likely pose the gravest potential security threat to South Korea. Even though maintenance of a cooperative economic environment of South Korea to both China and the US is stated as the political aim, the feasibility of such neutral management is questionable. As China attempts to expand its power in the West Pacific, it will have to clash with the US and its allies in a political and possible military conflict. In joint Press Statement in November 2017, Xi Jinping stated that “the Pacific Ocean is big enough to accommodate both China and the United States.”16 A military expansion to match such political aim would lead to a heightened chance of a conflict between the US and China that could impede South Korea’s high politics. For now, the clash between the US and China is far from being reminiscent of the Cold War. The economic links remain predominant between the two nations as the largest trading partners to each other render a degree of cooperation desirable for both parties, as indicated in a study by the South Korean Institute of Foreign Affairs and National Security. It is thus predicted to be the case for at least the next ten years, according to the same document. If this remains the case, South Korea may be able to secure a stable political position, thriving from the economic and political freedom of cooperating with the two of the largest powers in the world. The document also suggests that China has been primarily concerned with establishing political security in the region by means of economic development in cooperation with other nations, and it will likely prefer to maintain a friendly relation with other major Asian economies. This would be a reasonable assumption to be made as China owes its exponential economic growth to the diplomatic and economic opening up to the foreign markets as clearly evidenced by its more stabilised growth rates of national GDP in the years since 1978 when Deng’s reforms occurred, despite some exceptions of aggressive foreign policies (Sino-Vietnam War). South Korea needs friendly political relations with China for its cooperation to address North Korean issues. Despite efforts by South Korea, China, and the US, it has been difficult to achieve an effective and mutually agreed solution on North Korean denuclearisation. The US recognises this difficulty of achieving Chinese cooperation and compliance regarding the North Korean affairs if they are linked to other issues regarding the US-China bilateral relationship17. This North Korean case exemplifies the effect of the US and China failing to cooperate in achieving a crucial aim of South Korean high politics, namely North Korean denuclearisation. Military manoeuvres by South Korea and the US against North Korea have often been perceived to be a direct threat against China, as highlighted by the controversies surrounding the THAAD. The military manoeuvres’ political implications for China further complicate the relationship between the US, South Korea, and China, whereby it is difficult to maintain military security against North Korea without triggering Chinese concerns. However, the American political tension towards China is vital for South Korean security by preventing Chin a from offering overt support to North Korea. Even though it has been difficult to garner Chinese cooperation in isolati 15

Pak Michin, Kim Hwanghŭi. Korea National Diplomatic Academy, (www.ifans.go.kr/knda/ifans/kor/pblct/PblctView.do?pblctDtaSn=13570).

16U.S.

Mission China | 9 November, 2017 | Topics: News. “Remarks by President Trump and President Xi of China in Joint Press Statement: Beijing, China.” U.S. Embassy & Consulates in China, 24 Sept. 2018 china.usembassy-china.org.cn/selected-quotes-press-statement-president-trump-joint-press-conference-president-xi/. 17

“China's Role in North Korea Nuclear and Peace Negotiations.” United States Institute of Peace, 27 June 2019.

ng North Korea, the American contention against China rendered the process of isolation significantly more effective. The deterioration of North Korean security would be favourable for South Korea, not only to avoid North Korean aggr ession reminiscent of 1950, but also in its political leverage in negotiating with North Korea. The effectiveness of the p olicy of isolating North Korea in has been supported by the Ministry of Unification18 in citing the breakdown of the So viet influence and alliance as the turning point behind North Korea’s approach for maintaining communications and re lations with the US, South Korea, and Japan for a potential security solution. The Ministry of Unification also states th at the North Korean isolation has been consolidated by the Chinese and Russian reluctance in complete cooperation. T he American enforcement of such an agenda is, perhaps, the only feasible method of preventing North Korea from bei ng supported by China and Russia as opposed to the era when North Korea enjoyed support from the nascent People’s Republic of China or the Soviet Union in the 1950s. In a broader picture, avoiding involvement in a sphere of influence and establishing stable middle power diplomacy would be the aim of South Korea in securing its high politics, as illustrated by the Institute of Foreign Affairs and National Security19. In this context, a successful middle power diplomacy can refer to the policy of attempting to establish friendly relations with both the US and China without a clear commitment to one side as a smaller power. Such policy has been continuously favoured and exercised by past South Korean presidencies. The international chaos created by the COVID 19 pandemic provided an opportunity for South Korea to consolidate its international reputation as a favourable political partner. South Korea showed political support for China by not shutting its borders from China, but also provided vital aid to numerous nations including the US. The US Secretary of State spoke of the “assistance to the American people” as a testament to the “strong and unshakable” alliance between the nations. South Korea built a firm basis for future diplomacy to remain favourable to both superpowers while also expanding its influences on other nations through its assistance during the pandemic. This view is underscored by the Institute of Foreign Affairs and National Security: “The international community has witnessed how successfully South Korea has been coping with the pandemic crisis based on the principles of ‘openness, transparency, and democratic procedures.’ It means that people around the world obtained a positive perception of South Korea as trustworthy and reliable. And it could provide South Korea with strong momentum when it pursues a middle power diplomacy based on the universal, reliable, and trustworthy principles.”20 To reiterate, the political stance of South Korea is to maintain political bridges to both China and the US. As illustrated by Lee Hsien Loong, the Prime Minister of Singapore, “Asian countries see the United States as a resident power that has vital interests in the region. At the same time, China is a reality on Korea and Japan's doorstep.”21 Like other Asian nations, South Korea needs to consider the power of China despite its long-standing relationship with the US. It is in South Korea’s best interest to maintain its relations with both. However, a more active clash between the US and China may force South Korea into a situation whereby it needs to side with one, abandoning links with the other. Either choice would be devastating for South Korea, as illustrated by Moon Hee-Sang the speaker of South Korea’s legislature. In stating that “we cannot abandon economy for the sake of security, and we cannot abandon security for the sake of economy,” he highlights the intrinsic trade-offs related to both nations. South Korea guaranteed its security through close military cooperation with the US against neighbouring communist threats, especially during the Cold War. Nevertheless, drawing hasty parallels between the US-China clash and the Cold War would be a mistake as the military contrasts between the two conflicts repudiate such a comparison. The Soviet Union posed a serious threat to the US mainland with its expansive operating area, its huge nuclear arsenal, and a military rivalling that of the US. Even though the prospect of nuclear war appears to be far-fetched today, it was a real possibility in 1983; Dmitry Adamsky stated that the 1983 wargames “almost became a prelude to a preventative 18

“2017 puk'an ihae” [Understanding North Korea 2017] Ministry of Unification Institute for Unification Education (https://www.unikorea.go.kr/books/understand/understand/ebook/under_NK_2017/assets/contents/download.pdf).

Institute of Foreign Affairs and National Security is Korean government’s think-tank for foreign diplomacy and security

Minjeonghun. “COVID-19, Global Leadership, and the U.S.-China Relations.” Korea National Diplomatic Academy, 7 May 2020, www.ifans.go.kr/knda/ifans/kor/pblct/PblctView.do;jsessionid=odMTYtYi1FE-mfo0K6s80-P3.public21?pblctDtaSn=13580. 20

Ibid.

Loong, Lee Hsien. “The Endangered Asian Century”. 11 June 2020, www.foreignaffairs.com/articles/asia/2020-06-04/lee-hsien-loongendangered-asian-century.

nuclear strike”22. The same degree of military threat is posed by neither the Chinese nuclear capacity nor its conventional military power. The Chinese submarines are limited to the small areas of water surrounding China, and its military assets are surrounded by networks of American military bases and fleets around the Pacific. Thus, it is difficult to categorise this military clash a developing of a second Cold War, but rather a Chinese attempt of expansion against an asymmetric advantage in the American hands. China does not yet have the power to form a sphere of influence based on its military might. From the South Korean perspective, Cold War-like spheres of power will be the most undesirable, and this appears to be unlikely. However, when Xi Jinping expressed a desire for a political expansion, it is difficult not to imagine his desire for a simultaneous military expansion. His idea of sharing the Pacific Ocean with the US has resulted in confrontations over the South China Sea. By prohibiting the expansion of Chinese aggression and military activity, the US is arguably defending the sovereignty of nations, such as the Philippines, from falling powerless into a Chinese sphere of influence. South Korea admits that the gap between the Korean military and the Chinese one is too large to be contended with and is only increasing when considering the rate of China’s technological advancement. The Institute of Foreign Affairs and National Security lists the precise guided missiles able to reach Korea, effective unmanned fighters, naval expansion with multiple carrier fleets and submarines, information network, and the nuclear threat to be among the military threats posed by China to South Korea. Thus, the US provision of arms, technology, and training as an ally is welcomed by South Korea for its military security against China. The ROK armed forces have become one of the strongest militaries in the world, having grown by both internal developments out of necessity and the aid from the US in terms of technology, training, and supporting troops based in and around South Korea23. Even though there are controversies surrounding over-reliance on the US, close cooperation with the US bolstered South Korean ability to self-defend. The usage of American weapons and experiences of licence kickstarted the South Korean defence industry. For example, technologies for the K9 selfpropelled gun, exported to several nations, was expanded from the basis of the K55 (a variant of the American M109) and the development of the KAI T-50 Golden Eagle was accomplished through cooperation with Lockheed Martin. The successful development of the defence industry not only provided export revenues for the economy but also bolstered its security in the long-term as South Korea no longer relies on imports for most defence systems. It would also be wrong to neglect the American monetary aid in the early stages of the South Korean military’s development. A paper from the International Development and Cooperation Review highlights the significance of the US military aid in developing the South Korean military: “The rest of the maintenance costs and the entire spending on new weaponry and equipment to improve upon military strength are entirely dependent upon direct military assistance from the US.” 24 The defence support from the US alleviated the economic demand of upkeeping a large army while modernising it when South Korea lacked the budgetary freedom to do so. In fact, the paper refers to another data source, which suggests that the share of the US military aid in South Korean defence expenditures was much larger than the official South Korean defence budget, over 80% in the immediate post-war years and then gradually reduced to about 60% by 196025. Bearing in mind the incongruence with the official USAID and the unavailability of South Korean documents at the time, the American commitment tor the South Korean military was great. History reflects on the benefits of military cooperation with the US for establishing long-term military security. Abandoning such beneficial military ties would be a disaster, which has South Korea proactively worked to avoid for most of its history. Yet a question to be asked is whether this American influence on the South Korean military is beneficial in the long-run for the security of the nations in the face of the US-China conflict. The current deployment of over 23000 American soldiers in South Korea and defence systems, most notably the THAAD anti-missile defence system, leads to Chinese objection. The purpose of the THAAD, stated to be “defensive measure to ensure the security of ROK and its people, and to protect Alliance military forces from North Korea’s weapons of mass destruction and ballistic missile 22 Dmitry Dima Adamsky (2013) The 1983 Nuclear Crisis – Lessons for Deterrence Theory and Practice, Journal of Strategic Studies, 36:1, 441. 23

Chun, In-Bum. “Korean Defense Reform: History and Challenges.” Brookings, Brookings, 31 Oct. 2017, www.brookings.edu/research/korean-defense-reform-history-and-challenges/. 24

https://cpb-us-w2.wpmucdn.com/u.osu.edu/dist/2/20360/files/2017/04/GoR-final-w0lqqp.pdf25 Ham, 1998, p. 206.

Ham, 1998, p. 206.

threats” by the United States Forces Korea26, have been interpreted by China as a security threat that destabilises the entire region. China especially objects to the radar capacities of the THAAD system, which could detect missiles from within Chinese airspace, as well as its regional missile defence capacities ‘encircling’ China. Military decisions, as previously stated, are interlinked with geopolitical implications for the wider rivalry between the US and China. The THAAD was no exception, as China believed that its strategic interest in the region was undermined, even though the THAAD was installed with unfavourable settings in “terminal mode” and is “poorly positioned against Chinese intermediate-range missiles if launched at South Korea or Japan.”27 In retaliation, China banned tour groups from visiting South Korea and temporarily closed 75 Lotte stores in March 201728. The automobile and entertainment industries were also affected through unofficial sanctions. Moreover, there were local fears that the region may become a military target. The political and economic retaliation towards South Korea signifies a trade-off in the national security arising from the American military activity within the country, in contrast to the previous argument that American military influence bolsters national security. This may already be seen as a consequence of South Korea’s prioritisation of the US over China when neutral diplomacy could not be balanced. Japan’s security policy can symbolise what may become of the South Korean policy if it decided to prioritise its cooperation with the American military even more. Japan already stations two THAAD radars, and the USFJ (United States Forces, Japan) consists of 2000 army soldiers, 18000 marines, and 13000 sailors. Even though it is debatable if such reliance on the American forces would be necessary for South Korea with the continuation of modernisation and expansion of capacities of the ROK military, South Korean national security has a lot to gain from such a commitment should the need for it arises. CONCLUSION There is no clear answer for what the consequences of the US-China clash would be on the South Korean high and low politics. While South Korean low politics could benefit from increased competitiveness of industries in the global market as a result of the US-China trade war, there is also a risk of economic retaliations resulting from political decisions. Similarly, a thorough backing of the US may stabilise the South Korean high politics, but also increase the possibility of military conflict against a greater Chinese power. It is certain, however, that there will be unavoidable trade-offs when prioritising high politics over low politics or vice versa. It is easy to be preoccupied with the high politics threat posed to the very survival of the nation. It is perhaps understandable when much of South Korean history has been marked by existential threats. However, I believe that there is a modest chance of a military conflict between the US and China, at least for the imminent future. There is yet an urgent necessity to embrace the American alliance at the expense of economic and political relations with China. Security provided by a military alliance with the US is certainly welcome, and it will provide time for South Korea to consolidate its low politics in its economy and global reputation. However, if the US-China relationship becomes more contentious and poses a direct threat to South Korean national security, it should safeguard its high politics with an even stronger alliance with the US, while preparing for a new economic dynamic, as previously illustrated to be feasible and potentially beneficial in the long-run. Without the gravest of security threats to the imminent survival of South Korea, factors of the high politics and the low politics will always be interlinked, and must be appropriately considered from the South Korean perspective while navigating various international relations.

26ROK

& U.S. Joint Statement: ROK-U.S. Alliance Agrees to Deploy THAAD. 7 July 2016, www.usfk.mil/Media/News/Article/831175/rok-usjoint-statement-rok-us-alliance-/.

“THAAD on the Korean Peninsula.” Institute for Security and Development Policy, 9 Oct. 2017, isdp.eu/publication/korea-thaad/.

Ibid.

Military Cooperation with South Korea and Japan: What the United States Wants, but Cannot Easily Achieve In Yeol Yeo Daewon Foreign Language H.S. in the Republic of Korea dannydream02@gmail.com

I. INTRODUCTION The Northeast Asian region has always been one of the most important stages for the United States in terms of expanding global influence through trade and military presence. In the region, there are countries such as the Russian Federation, the People's Republic of China, the Republic of Korea, the Democratic People’s Republic of Korea and Japan. However, this paper will only focus on the following three countries: China, South Korea, and Japan. These countries are especially affected by the ongoing conflict between the United States and China. Following the commencement of the Trump administration, the Northeast Asian region has especially been turbulent, with the United States playing a major role in the changes. Especially after 2017, China has been one of the main concerns of the United States—it was designated as the “strategic competitor.” This change can be found in numerous government documents and laws concerning China. At the same time, the United States has gradually turned to its two main allies in the region, South Korea and Japan. These two countries seem to be assuming more important roles in economic and military partnership with the United States. The National Defense Authorization Act, an annual federal budget law for the Department of Defense (DOD), well reflects the direction of strategies or relations that the United States hopes to take or establish.1 Especially, policies related to military and security are effectuated through this Act, since it is one of the vehicles through which the Congress authorizes what the Department of Defense can or cannot do.2 However, despite its importance, scholars in relevant fields seem to have underestimated the importance of the NDAA. The Act deserves more attention from academia. Thus, by analyzing the 2017, 2018, 2019 and 2020 NDAAs, this paper will first show that the United States wants more military cooperation with Japan and Korea as US-China conflicts escalate. Yet, through detailed analysis of the Korean and the Japanese governments’ reactions to the request of the United States, this paper will also tell the readers that the United States will find it difficult to make the military cooperation happen as robustly as it wishes. II. HISTORICAL REVIEW OF THE US-CHINA RELATIONS The United States has never lost its clout over the Northeast Asian Region after World War II. In short, the relations between the United States and China were mostly the repetition of normalization and deterioration depending on the administrations of both countries. This can be even seen in the recent decade of the U.S. – China relations. During the Obama administration, the relationship between China and the U.S. improved to a certain level. The administration established the U.S.–China Strategic and Economic Dialogue in 2009 to discuss cooperative actions in solving economic crisis and environmental problems. 13 While Obama’s “Pivot to East Asia” in 2012 actually did increase military tension or pressure against China in the Northeast Asian region, the office tried to find a balance between pressure and partnership with the Eastern giant. 4 The office greatly supported Xi Jinping’s “New Model of Great Power Relations” and also the “One-China” policy, taking sides with China in global powers or some sensitive political matters. 5

1,2 Erickson, Andrew S. “U.S.-China Military-to-Military Relations: Policy Considerations in a Changing Environment.” Asia Policy, vol. 26, no. 3, July 2019, pp. 123–144., doi:10.1353/asp.2019.0041. 3,4,5

Green, Michael J. “The Legacy of Obama's ‘Pivot’ to Asia.” Foreign Policy, Foreign Policy, 29 May 2019, foreignpolicy.com/2016/09/03/the-legacy-of-obamas-pivot-to-asia/; Council on Foreign Relations. “Timeline: U.S. Relations With China 1949– 2020.” Council on Foreign Relations, Council on Foreign Relations, www.cfr.org/timeline/us-relations-china

However, since the Trump administration, relations between the two have greatly deteriorated. On August 2017, President Trump, through the United States Trade Representative's invocation of Section 301 of the 1974 Trade Act, conducted an extensive investigation on the unfair trade and transfer of technology perpetuated by the Chinese government.2 China was found out to be using the following means: 1) “Foreign owner restrictions, including joint venture requirements, equity limitations, and other investment restrictions” 2) “Substantial restriction on, and intervention in U.S firms’ investments and activities” 3) “Directing and facilitating the systematic investment in, and acquisition of, U.S. companies and assets by Chinese companies” 4) “Conducting and supporting unauthorized intrusions into, and theft from, the computer networks of U.S. companies” Along with these problems and an unprecedented trade deficit with China, President Trump initiated strong economic sanctions on China from March 2018. He implemented them through imposing tariffs and restricting certain Chinese investments on the United States soil. As a response, the Chinese government also imposed major retaliatory sanctions on the United States products to worsen the relations between the two.3 This hostile relation triggered by a trade war is now expanding to national security issues. In 2018 and 2019, certain Chinese telecommunications companies such as Huawei and ZTE were added on the Entity list of the United States Bureau of Industry & Security; they were accused of transferring sensitive United States technology to the Chinese government.4 The United States have interpreted such espionage of companies as an active extension of the Chinese government’s “Made in China 2025” policy, which is a movement by the Chinese government to change the economic structure from labor-intensive to high-tech industry. The U.S. concluded that the actions were backed by the government rather than being a single company’s separate action.5 Christopher Wray, the FBI director, called China the "greatest long-term threat to our nation's information and intellectual property and to our economic vitality" and noted the FBI has more than 2,000 open investigations that tie back to the Chinese government.6 As a result, the whole diplomatic relations between the U.S. and China has certainly worsened. On July 23rd 2020, Mike Pompeo, the Secretary of State of the United States, strongly denounced the CCP (Chinese Communist Party) for its violation of human rights, perpetuation of unfair trade, and usurping of global order. He even urged the people in China to rise up against the CCP.7 Robert O’Brien, the White House national security advisor, has criticized what he calls the Chinese Communist Party’s totalitarian vision and said “Let me be clear — we have deep respect and admiration for the Chinese people. The United States has a long history of friendship with the Chinese nation. But the Communist Party does not equal China or her people.”8 Although the growing power of the Chinese military and its nuclear weapons was always a concern, China is now considered the major threat and a main opponent against the national security of the United States. It is no longer a military or an economic partner. III. ANALYSIS ON NATIONAL DEFENSE AUTHORIZATION ACTS 1. What is NDAA

Office of U.S. Trade Representative. President Trump Announces Strong Actions to Address China's Unfair Trade. 22 Mar. 2018, ustr.gov/about-us/policy-offices/press-office/press-releases/2018/march/president-trump-announces-strong.

Timmons, Heather. “Timeline: Key Dates in the U.S.-China Trade War.” Edited by Leslie Adler, Reuters, Thomson Reuters, 16 Jan. 2020, www.reuters.com/article/us-usa-trade-china-timeline/timeline-key-dates-in-the-u-s-china-trade-war-idUSKBN1ZE1AA.

Bureaus and Offices, “Department of Commerce Announces the Addition of Huawei Technologies Co. Ltd. to the Entity List.” U.S. Department of Commerce, 2 Aug. 2019, www.commerce.gov/news/press-releases/2019/05/department-commerce-announces-addition-huaweitechnologies-co-ltd. 5 Office of U.S. Trade Representative. President Trump Announces Strong Actions to Address China's Unfair Trade. 22 Mar. 2018, ustr.gov/about-us/policy-offices/press-office/press-releases/2018/march/president-trump-announces-strong. 6

Shortell, David. “FBI Director Unleashes on China in Speech.” CNN, Cable News Network, 7 July 2020, edition.cnn.com/2020/07/07/politics/christopher-wray-fbi-director-china-speech/index.html.

Delaney, Robert. “Pompeo Urges China's Citizens to Help Change Beijing's 'Behaviour'.” South China Morning Post, 24 July 2020, www.scmp.com/news/china/diplomacy/article/3094467/us-secretary-state-urges-chinas-citizens-help-change-behaviour.

Allen-Ebrahimian, Bethany. “O'Brien Says China's Global Influence Puts Americans at Risk.” Axios, 24 June 2020, www.axios.com/robertobrien-china-influence-americans-risk-c01a3998-be33-4d8f-9e6c-b60a9185a904.html.

National Defense Authorization Act, or the NDAA, is the United States federal law that specifies and sets policies, and programs for the U.S Department of Defense for each fiscal year.13 Specifically, the NDAA contains not only plans for domestic affairs, but also matters related to foreign nations such as maintaining deployed U.S troops, performing joint military training, and establishing strong security around the global regions.14 2. Scope of research & significance of NDAA in the US -Northeast Asia dynamics In the National Defense Authorization Act, (NDAA) the Northeast Asian region has always been one of the most important strategic points. Especially the five major countries – Russia, China, South Korea, North Korea and Japan have been continuously mentioned in the Act. However, starting from the Trump administration, the portion of the law’s focus on China, South Korea, and Japan has significantly increased. The NDAA, which deals with matters relating to foreign nations as mentioned above, will well reflect the direction of the military or security strategies that the U.S. hopes to take toward the three countries in view of the US – China conflict. Thus, this paper will first briefly discuss the NDAA 2017 to understand the initial position of the United States on its alliance with Japan and South Korea prior to the commencement of the Trump administration. Then, it will conduct a comprehensive analysis of the NDAA 2018, 2019, and 2020 to understand the direction that the U.S. encourages the Republic of Korea and Japan to take toward the US – China conflict. Understanding the directions, this paper will then compare it with the responses of Japan and South Korea. 3. National Defense Authorization Act 2017 a) U.S. – China relations The 2017 National Defense Authorization Act, which was passed under the Obama administration, did not strongly focus on directly pressuring China. Instead, it only used limited indirect measures to show concern for the growing power and military activities in China. In the section 1271 of the NDAA 2017, it showed that the Act only requires the Department of Defense to submit an “Annual report on military and security developments involving the People’s Republic of China.”15 Moreover, other sections in the NDAA 2017 suggested that the main concern and focus of the United States for the fiscal year 2017 were matters regarding the Russian Federation, the DPRK, and terrorism in the Middle East.16 Overall, looking at the NDAA 2017 in a holistic view, issues regarding the foreign nations were mostly focused on countries other than China. Even if there were some matters related to China, the Act did not require strong and direct measures to be taken toward China. Rather, it simply expressed concern about its growing military power in the Northeast Asian region. b) The role of Japan and South Korea in the U.S. – China conflict In the NDAA 2017, where China was not yet viewed as a main enemy or an opponent in the Northeast Asian region, the roles of Japan and South Korea also had yet very little to do with China. Rather, each country was only emphasized as an individual ally of the United States. In the Act, the direct reference of Japan only appeared in several joint military construction projects with the United States. For instance, section 2201 “Authorized Navy construction and land acquisition projects” contained a subsection allowing the Department of Defense to spend 42 million dollars for construction in two Japanese bases: Kadena Air Base and Sasebo.17 In the case of the Republic of Korea, the situation was similar to that of Japan. Direct reference of South Korea as an individual was only mentioned in the section 2511 “Republic of Korea funded construction projects.”18 In the section, the Act allowed the DOD to conduct military constructions in the Republic of Korea worth of 616.8 billion 13,14

“S.2943 - 114th Congress (2015-2016): National Defense Authorization Act for Fiscal Year 2017.” Congress.gov, 23 Dec. 2016, www.congress.gov/bill/114th-congress/senate-bill/2943. 15,16,17

United States, Congress, Cong. Senate, Armed Services. National Defense Authorization Act for Fiscal Year 2017: Conference Report to Accompany S. 2943, U.S. Government Publishing Office, 2016. 114th Congress, 1st session, bill S.2943.

18,19,21

dollars. Overall, the Act did not go anywhere further, as it did not mention any specific roles for both the U.S. -Japan and U.S. – South Korea alliance in the US – China conflict. The trilateral cooperation of the three countries was, in fact, mentioned in the NDAA 2017. It was mentioned from the expression “Sense of Congress on trilateral cooperation between Japan, South Korea, and the United States” in the Act.19 However, there were two points why the mentioning of the ‘trilateral cooperation’ was relatively insignificant in the NDAA 2017. First of all, the expression was under the portion “Legislative provision that is not adopted,” meaning that it had not been approved by the Senate and the House of Representatives to become an official section in the NDAA. This did not mean that trilateral cooperation was not made in the year of 2017. The three countries did hold the “9th Defense Trilateral Talks” on April 19, 2017.20 However, what this shows is that the ‘trilateral cooperation’ was not emphasized greatly or interpreted as significant in the NDAA 2017. Secondly, the expression also focused on the role of the three countries in security matters only related to the 21 DPRK. Under the expression, there was not a direct reference of any country except the DPRK, and there wasn’t any implication on issues regarding China. Even in the actual “Joint Statement of the 9th Defense Trilateral Talks,” China was never mentioned as a great concern.22 The reason behind these two points is likely to be that China had not yet been seen as the main ‘enemy’ in the Northeast Asian region. Overall, it can be explained that the roles of Japan and South Korea in combating China alongside the United States were not emphasized at all in the NDAA 2017. c) Summary The NDAA 2017 showed the initial stance of the United States on Japan and South Korea’s role in the Northeast Asian region prior to the establishment of U.S. – China conflicts. In the Act, Japan and South Korea were only referred to as individual allies of the United States in the region. The trilateral cooperation itself was barely mentioned in the Act, and even in the small reference, it focused only on the role of promoting peace in the Korean peninsula. As the Sino-American relations had not yet become hostile, the role of Japan and South Korea was more encouraged towards keeping peace in the Korean peninsula. 4. National Defense Authorization Act 2018, 2019 and 2020 a) U.S. – China relations The NDAA 2018, 2019 and 2020 clearly showed the strengthening of both indirect and direct measures of the United States on China. For instance, the increasing emphasis on matters relating to Huawei and ZTE showed growing concern on China as a national security threat.23 A stronger requirement of the DOD to submit a more detailed, narrowed down “Annual report on military and security developments involving the People’s Republic of China,” also represented the strengthened will of the U.S. to monitor China as an enemy now.24 b) Increased cooperative role with Japan and South Korea In the NDAA 2018, 2019, and 2020, the United States started to further encourage military cooperation from South Korea and Japan as the US – China relations deteriorated. The most enticing part is the insertion of “Subtitle F- Matters relating to Indo-Asia-Pacific Region.”25 Under the new subtitle, there were phrases that implicitly refer to China as a serious threat to the security of the East Asian region. As a response, the United States aimed to expand the military roles of Japan and South Korea from the Korean peninsula to address the challenges posed by China. i) Maintaining U.S. troops in the countries

20,22

Joint Statement of the 9th Defense Trilateral Talks. dod.defense.gov/Portals/1/Documents/pubs/Joint_Statement_of_the_9th_Defense_Trilateral_Talks_April_19_2017.pdf?ver=2017-04-19104321-710. 23,24,25

These appear on all three versions of the NDAA: 2018, 2019, and 2020; United States, Congress, Cong. House, Armed Services. National Defense Authorization Act for Fiscal Year 2018: Report of the Committee on Armed Services, House of Representatives, on H.R. 2810, Together with Additional Views (Including Cost Estimate of the Congressional Budget Office), U.S. Government Publishing Office, 2017. 115th Congress, 1st session, bill H.R.2810; United States, Congress, Cong. House, House - Armed Services. John S. McCain National Defense Authorization Act for Fiscal Year 2019, U.S. Government Publishing Office, 2018. 115th Congress, 1st session, bill H.R 5515.; United States, Congress, Cong. Senate, Armed Services. National Defense Authorization Act for Fiscal Year 2020: Conference Report to Accompany S. 1790, U.S. Government Publishing Office, 2019. 116th Congress, 1st session, bill S.1790

Compared to the previous NDAAs, starting from the NDAA 2019, the importance of the U.S. troops in foreign bases, especially those in South Korea and Japan, was reaffirmed and emphasized. Under the added section 1255 “Sense of Congress reaffirming security commitments to the governments of Japan and South Korea and trilateral cooperation between the United States, Japan, and South Korea,” the United States highlighted its commitment to the Mutual Defense treaties signed with Japan and South Korea.26 Specifically, the NDAA 2019 expressed the need to strengthen the alliance regarding South Korea by adding a new section 1264 “Limitation on Use of Funds to Reduce the Total Number of Members of the Armed Forces Serving on Active Duty who are Deployed to the Republic of Korea.”27 After the signing of the Status of Forces Agreement (SOFA) to deploy U.S. troops in Korea in 1966, the United States had paid the entire cost for the United States Forces Korea (USFK) until 1991.28 However, since 1991, a Special Measures Agreement (SMA) has been introduced, resulting in both countries sharing the defense cost.29 Ten SMAs have been signed until the present, with the last one having expired on December, 2019. An 11th agreement has not been yet settled due to President Trump’s demand for a 5 fold increase from the previous agreement.30 He has even threatened to withdraw the United States Force Korea (USFK) if this demand is not met.31 From the newly added section 1264, we can infer that there was a gap in positions regarding USFK between the executive and legislative branch of the United States. In concern of Trump’s threat, the Congress noted that “None of the funds authorized to be appropriated by this Act may be used to reduce the total number of members of the Armed Forces serving on active duty who are deployed to the Republic of Korea below 22,000.”32 It reiterated the importance of USFK in the Northeast Asian region. However, importance of these statements is driven home by looking at how this section was placed under the subtitle F – “Matters Relating to the Indo-Pacific Region.” It can be interpreted that the Congress recognized the USFK’s role of not only promoting peace in the Korean peninsula but also pressuring China. In fact, the presence of the USFK is often depicted as a “dagger” to China and a key U.S. tool in rebalancing peace in the Asian region vis-à-vis China’s rise.33 The number of the USFK was increased to 28,500 in the NDAA 2020, which means the Congress felt increased importance as the US – China relations worsened.343 Overall, the role of South Korea and Japan in the U.S. – China conflict was first expressed as frontiers maintaining the U.S troops in the Northeast Asian region. ii) Installation of Military Equipment From the NDAA 2018, the installations of military equipment, especially long-range surveillance radars, had been more emphasized with respect to the roles of South Korea and Japan. In the NDAA 2017, “strong support” for the deployment of THAAD in South Korea was mentioned briefly.34 Based on the fact analysis that THAAD “could improve early tracking of some Chinese missiles,” it can be presumed that THAAD could had acted also as a means to monitor China’s

26,27

These specifically refer to NDAA 2019 and NDAA 2020.

United States, Status of Forces Agreement, 1966, usfk.mil, https://www.usfk.mil/Portals/105/Documents/SOFA/A01_SOFA.Art.I-XXXI.pdf

United States, Special Measures Agreement, 1991, state.gov, https://www.state.gov/wp-content/uploads/2019/06/19-405-Korea-SMA.pdf

Park, Boram. “S. Korean Defense Minister Denies Bolton's Claim on Defense Cost-Sharing Deal.” Yonhap News Agency, 24 June 2020, en.yna.co.kr/view/AEN20200624008700315.

31 Stracqualursi, Veronica. “Trump Apparently Threatens to Withdraw US Troops from South Korea over Trade.” CNN, Cable News Network, 16 Mar. 2018, edition.cnn.com/2018/03/15/politics/trump-us-troops-south-korea/index.html. 32

This specifically refers to the NDAA 2019.

33 Lee, Seong-Hyon. “The Shift of Security Environment in Northeast Asia: The US–China Conflict and Its Implications for Korea.” Journal for Peace and Nuclear Disarmament, vol. 1, no. 2, 2018, pp. 352–362., doi:10.1080/25751654.2018.1542655. 34

This specifically refers to the NDAA 2020.

missile bases along its East Coast.35 However, the way how the NDAA 2017 only mentioned the DPRK as the target country for the THAAD technology shows the Obama administration’s equivocal attitude towards China.36 In the NDAA 2018, the United States enlarged the limited scope of THAAD from targeting DPRK to include China, well representing the Trump administration’s explicit hostility to the country.37 Moreover, the Act encouraged the installation of further ballistic missile defense mechanisms in South Korea and Japan.38 Therefore, it can be said that the United States considered South Korea and Japan as important strategic points to deploy additional military equipment in the US – China conflict. iii) Trilateral Cooperation In the NDAAs for2018, 2019, and 2020, the importance of trilateral cooperation was also emphasized. Starting from the NDAA 2018, under section 1255, the United States sought to carry out “expanded exercises, training, and information sharing that strengthen integration,” and “greater security cooperation with, and among, Japan and South Korea.”39 Different from the NDAA 2017, the section encouraged the trilateral cooperation to secure peace not only on the Korean peninsula but also in the whole Indo-Asia Pacific region following the worsening US – China relations.40 This shows that the United States started hoping for a cooperative participation of both countries in pressuring China. c) Summary of the analysis of NDAAs Recalling the NDAA 2017, the main concerns of the U.S. were more towards issues rising in Russia, Syria and DPRK, not China. However, from the NDAA 2018, the Trump administration’s first NDAA, the direction of the Acts had sharply shifted towards targeting China. Due to the increasing hostility towards China, the United States also emphasized and encouraged the roles of South Korea and Japan in combating China in three ways : 1) maintaining U.S. foreign troops , 2) installing additional military equipment, and 3) forming trilateral cooperative actions or treaties. The NDAAs from the 2018 viewed South Korea and Japan as the “cornerstones for the preservation of peace and stability in the IndoAsia-Pacific region.” It shows that the two allies were viewed necessary in the U.S. – China conflict. IV. THE RESPONSES OF JAPAN AND THE REPUBLIC OF KOREA ON COOPERATION TO THE USCHINA CONFLICT 1. Connection to the NDAAs The analysis of the NDAAs 2018, 2019, and 2020 shows that the United States encouraged the participation of Japan and South Korea in its increasingly tense interactions with China. In this section, this paper hopes to answer these questions: Were efforts to secure further participation from Korea and Japan successful ,and, to the extent they were, what made them work? If not, what kind of barriers were there? The answers to these questions will provide the direction that the United States should pursue to reach a better cooperation.

Maintaining U.S. troops in the countries In the case of South Korea, there have been number of issues that have impeded cooperation. South Korea is trying to lose its dependency on the United States military. Though the intention of the South Korea’s policies may not be directed towards eliminating the cooperative spirit emphasized by the United States, it seems that some portion of it may inevitably be undermined. There are a couple of major military issues that seem to negatively affect the relationship between South Korea and the United States in terms of security. The first is the matter of the ongoing negotiation on renewing the Special Measures Agreement (SMA). The second is the matter of Wartime Operational Control, which the United States currently has but South Korea seeks to retrieve.

Panda, Ankit. “What Is THAAD, What Does It Do, and Why Is China Mad About It?” – The Diplomat, For The Diplomat, 26 Feb. 2016, thediplomat.com/2016/02/what-is-thaad-what-does-it-do-and-why-is-china-mad-about-it/.

This specifically refers to the NDAA 2017.

37,38

These specifically refer to the NDAA 2018.

39,40

These appear on all three versions of the NDAA: 2018, 2019, and 2020.

President Moon’s response is that South Korea is willing to increase its defense cost sharing by an incremental percentage increase like it has been for the past 30 years, but cannot accept a 400% increase.41 An interesting point is that as South Korea is still in dispute with Washington regarding the defense cost sharing, it is building a closer relationship with China. Witness the 2018 Defense White Paper of Republic of Korea which has the ROK and PRC increasing military cooperation and promoting frequent meetings of high officials.42 This movement cannot be portrayed as favorable to the United States. Such responses definitely weaken the cooperative spirit between the U.S. and South Korea, since China has become the United States’ hostile counterpart. Meanwhile, President Moon’s mentioning about Wartime Operational Control comes into light. Wartime Operational Control is an “authority to perform functions of command over subordinate forces.”43 In South Korea, when a war occurs, the authority will be granted to the Commander of the United States Forces in Korea, instead of an official of the South Korean military.44 President Moon has been arguing for the transfer of this Wartime Operational Control over to South Korea’s military even before becoming the president, and now the specific plans are explained in the 2018 Defense White Paper of Republic of Korea.45 What makes it worse for the United States is that South Korea is even trying to pursue complete ‘self-defense,' someday removing the USFK from the peninsula.46 For South Korea, this is reasonable because in the defense cost sharing between the two countries, 46% of the whole cost has been being used in construction and housing fees for the USFK forces, not practically on expanding the armed powers.47 If South Korea is willing to use such funds for its own military development, ‘self-defense’ might be more practical for South Korea. The two current issues seem to work against the cooperative spirit between the United States and South Korea. With South Korea gradually detaching itself from the USFK, if President Trump’s threat to pull out the USFK continues, it might actually be South Korea that could remove USFK from its territory. This is definitely not what the U.S. would wish to occur in the middle of a US – China conflict. Unfortunately, this goes the same with Japan. Officials from the Japan Ministry of Foreign Affairs explained the 8 billion-dollar demand in defense sharing cost as “nearly impossible.”48 The country’s self-defense forces has been described as a “toothless tiger” without the help of the United States.49 However, the Japan Self-Defense Forces (JSDF) is changing in terms of characteristic and power. Ever since it was created in 1954 with the help of the U.S., soon after the “Security Treaty Between the United States and Japan” in 1951, it has been expanding its power.50 Article 9 of the Japanese Constitution still prohibits the forces from conducting war, developing or procuring any offensive weapons.51 However, in the 21st century, this sole defensive characteristic of the JSDF has started to change. In 2007, the section 2 of Article 3 of the Self Defense Forces Act was amended, stating that abroad activities of

Park, Boram. “S. Korean Defense Minister Denies Bolton's Claim on Defense Cost-Sharing Deal.” Yonhap News Agency, 24 June 2020, en.yna.co.kr/view/AEN20200624008700315.

42,43,44

These all refer to the 2018 Defense White Paper of Republic of Korea. Defense White Paper of Republic of Korea is the Korean version of the NDAA, which is published biennially; Republic of Korea, Ministry of National Defense, 2018 Defense White Paper, Ministry of National Defense, 31 Dec. 2018, https://www.mnd.go.kr/user/mnd/upload/pblictn/PBLICTNEBOOK_201907110548253080.pdf

This also specifically refers to the 2018 Defense White Paper of Republic of Korea.

Shihyeon, Jo. “문재인 대통령 ‘자주국방 힘 위에서 한미동맹 굳건해진다.’” Newbc 뉴비씨, 19 July 2019, www.newbc.kr/news/articleView.html?idxno=6170.

This also specifically refers to the 2018 Defense White Paper of Republic of Korea.

48 Lee, Hae-Young, 일, 미방위분담금 5 배 증액 요구에 ‘곤혹’…”불가능에 가깝다” [‘Frustration’ of Japan with 5 fold increase in request of defense cost sharing by the U.S. … “Nearly impossible”], Yonhap News, 2019. https://www.yna.co.kr/view/AKR20190801062800073 49 Gady, Franz-Stefan. “Toothless Tiger: Japan Self-Defence Forces.” BBC News, BBC, 15 Oct. 2015, www.bbc.com/news/world-asia34485966. 50

Stewart, Taylor. “Does Japan Have an Army?: Culture: Metropolis Magazine Japan.” Metropolis Japan, 9 Aug. 2019, metropolisjapan.com/culture-japanese-self-defense-forces/. 51 Stewart, Taylor. “Does Japan Have an Army?: Culture: Metropolis Magazine Japan.” Metropolis Japan, 9 Aug. 2019, metropolisjapan.com/culture-japanese-self-defense-forces/.

the JSDF are “basic duties,” not “miscellaneous regulations” like in the past Acts.52 Starting from this amendment, the JSDF started to incorporate some offensive traits, allowing the forces to conduct further foreign campaigns. The Japanese government uses this as a justification to strengthen the forces. In terms of power and the support of the government, the JSDF should never be ignored. In 2011, Japan had the world’s eighth-largest military budget, and it has been continuously rising under the Abe administration.53 In 2020, Japan is ranked as the 5th most powerful military force on the planet - a ranking hard to explain for a ‘Self-Defense Forces.’54 Japan no longer has only a Self-Defense Force, but, rather, an army capable of not only protecting its own territory but also threatening the other nations. It is no longer a country that practically needs the U.S. to protect its territory. With this current trend of the JSDF, it again seems like the situation is not favorable for the United States. If President Trump’s “nearly impossible” demand for increase in defense sharing costs and threats to remove the USFJ (United States Force Japan) continues, it might be Japan which is more willing to push the United States out of its territory. There is not much to lose for Japan now. With less spending on defense sharing, the country can promote a stronger military of its own, and it would even better allow Abe Shinzo to amend the Article 9 of the Japanese constitution with less intervention from the United States. In this situation, the one who wants to keep the U.S. foreign forces in Japan and South Korea is really the United States government. However, it is the U.S. that is burning the candle from both sides. The current government of South Korea is generally not supportive of pressuring China with the USFK. Moreover, for South Korea and Japan, there is not much merit to maintain the USFK, or USFJ with a greater cost than now. With President Trump strongly pushing his demand while the two Asian countries are reasonably reluctant to sign to this massive increase, there are possibilities that they might take abrupt routes that the United States never thought of – the removal of U.S. foreign forces.

Installation of Military Equipment For the ‘installation of military equipment,’ it is also not too much different from the first point mentioned above: the responses of the two allies are not really harmonious with the United States. In South Korea, Terminal High Altitude Area Defense (THAAD), one of the military equipment mentioned in the NDAAs, was first deployed to South Korea in 2017 under the Park Keun Hye’s administration.55 It was marked as a sign of long-lasting alliance between the U.S. and South Korea. South Korea overcame not only the domestic opposition at the installation sites, but also opposition from China; it expressed great agitation towards the United States’ deployment. The reason behind this criticism was that THAAD will be used to monitor the Chinese military sites.56 However, from the extension of President Moon’s support for the ‘self-defense’ of South Korea, he has continuously opposed major installations of military equipment inside the Korean peninsula. Especially, in 2017, President Moon was the one who strongly denounced the decision of installation, calling it an “undemocratic decision” before becoming the president.57 After elected as the president, although he temporarily allowed the initial deployment 52

The Constitution and the Right of Self-Defense. www.mod.go.jp/e/publ/w_paper/pdf/2007/21Part2_Chap1_Sec2.pdf; Sieg, Linda. “Mission Unaccomplished - Abe's Drive to Revise Pacifist Constitution.” The Japan Times, 19 Nov. 2019, www.japantimes.co.jp/news/2019/11/19/national/politics-diplomacy/shinzo-abe-revise-constitution/.

Stalenheim, Petter. “Military Expenditure” SIPRI Yearbook 2008: Armaments, Disarmament and International Security. Jan. 2019. https://www.sipri.org/sites/default/files/YB08%20175%2005.pdf

54 Global Firepower, “Japan Military Strength (2020).” Global Firepower - World Military Strength, www.globalfirepower.com/countrymilitary-strength-detail.asp?country_id=japan. 55

“US and South Korea Agree THAAD Missile Defence Deployment.” BBC News, BBC, 8 July 2016, www.bbc.com/news/world-asia36742751.

Glosserman, Brad. “Seoul Draws Wrong THAAD Lessons.” The Japan Times, 27 Jan. 2020, www.japantimes.co.jp/opinion/2020/01/27/commentary/japan-commentary/seoul-draws-wrong-thaad-lessons/. 57 Journal, The Asia Pacific. “Moon Jae-In's THAAD Conundrum: South Korea's ‘Candlelight President’ Faces Strong Citizen Opposition on Missile Defense.” The Asia-Pacific Journal: Japan Focus, 15 Sept. 2017, apjjf.org/2017/18/Martin.html; Sang-hun, Choe. “South Korean Villagers Protest Plans for U.S. Missile Defense System.” The New York Times, The New York Times, 13 July 2016, www.nytimes.com/2016/07/14/world/asia/south-korea-thaad-us.html.

of two THAAD launchers, he strongly demanded an extensive investigation on the hidden four additional launchers that the Korean Defense Ministry had planned to install.58 Moreover, again looking at the 2018 Defense White Paper of Republic of Korea, the Moon administration acknowledged Chinese government’s concern regarding THAAD. It stated how the Korean government continuously tried to mitigate this tension by expanding bilateral cooperation with China means such as pursuing stronger “defense exchanges, cooperation in a more substantive manner by revitalizing senior official exchanges and establishing exchanges among military units…”59 These efforts can be disappointing for the United States, as it would not want South Korea to form deep military ties with China. For the United States, the current administration of Korea can be interpreted as a less cooperative one, undermining the integrity of the long-lasting alliance between the two. Therefore, it can be predicted that installing other means of defense mechanisms would not be easy as the two countries would not reach a swift agreement. South Korea would not want to risk its economic connection with China for installing further defense mechanisms. This is very reasonable because since 2007, China has always been Korea’s largest trading partner, and now takes more than 20 percent of the whole trade of South Korea.60 In 2017, when China banned tourism to Korea and certain Korean industries in the mainland China after the deployment of THAAD, the Republic of Korea had to pay a price of $7.5 billion.61 This was almost 0.5 percent of its GDP.62 Based on such facts, it is hardly likely for the Republic of Korea to easily cooperate on the plan of the United States on installing further military equipment. For South Korea, it must avoid another dispute with China. If South Korea is to make another decision unfavorable to Korea-China relations, it would have to face another catastrophic economic loss in terms of economic revenue from and retaliations imposed by China. For Japan, discussions on military equipment have already taken place, regarding installation of Aegis Ashore, and the end result was not satisfactory for the U.S. The Japanese government saw insufficient reason to install such equipment with great costs and domestic oppositions. There were ongoing talks to deploy two Aegis Ashore, another missile defense system created by the Lockheed Martin Corporation, in the Akita and Yamaguchi region from 2017.63 However, on June 2020, the National Security Council of Japan confirmed that the decisions have been suspended.64 With the great cost for each launcher and the massive protests from the people in Akita and Yamaguchi, worrying that they would be first targets in case of North Korean threats, the Japanese government really had no choice but to suspend the talk. For the United States, certain military equipment in Japan and South Korea can serve as great monitoring devices on Chinese military. However, again it seems like it is only the United States that is greatly anxious under this situation. South Korea is not likely to risk its ties with China by cooperating in military equipment installations. Japan, which is supportive of the idea itself, is now withdrawing from the plan because of practical problems: cost and domestic opposition. Another goal in the NDAAs is starting to appear unlikely for the U.S.

Trilateral Cooperation When the NDAAs of 2018, 2019, and 2020 encouraged trilateral cooperation among the U.S., Japan, and South Korea, they intended cooperation that directly involved ‘all’ three countries. As mentioned in the NDAAs, the three countries signed a trilateral intelligence sharing agreement on 2015, whereby the U.S. Secretary of Defense would regularly meet with the officials from both the South Korea’s Ministry of National Defense and Japan’s Ministry of Defense.65 There are many other “small-scale trilateral and tabletop exercises,” but the most noticeable trilateral cooperation is the 58

Reif, Kingston. “Arms Control Today.” Arms Control Association, Sept. 2017, www.armscontrol.org/act/2017-09/news/moon-reverses-thaaddecision.

This specifically refers to the 2018 Defense White Paper of Republic of Korea.

60,61,62

Glosserman, Brad. “Seoul Draws Wrong THAAD Lessons.” The Japan Times, 27 Jan. 2020, www.japantimes.co.jp/opinion/2020/01/27/commentary/japan-commentary/seoul-draws-wrong-thaad-lessons/.

63,64

Joint Statement of the 9th Defense Trilateral Talks. dod.defense.gov/Portals/1/Documents/pubs/Joint_Statement_of_the_9th_Defense_Trilateral_Talks_April_19_2017.pdf?ver=2017-04-19104321-710.

GSOMIA, or the General Security of Military Information Agreement between South Korea and Japan. It is officially a bilateral intelligence sharing pact, but it has acted as the key dimension in the ‘trilateral cooperation.’ It strengthened the Korea - Japan connection, which was always the weakest point in the triangle. For that reason, the United States has strongly supported the GSOMIA ever since it was established, and urged the two allies to share a great bound of information to secure peace in the Indo-Asia Pacific region.66 However, the crucial Korea – Japan connection is crumbling for the United States. Although the relations between the two have long been problematic, given their history and more recently resulting from sovereignty issues over the Dokdo (what Japan calls ‘Takeshima’) in the East Sea, the tensions between the two reached its peak in 2018. The South Korean Supreme Court’s decision requiring the Japanese companies to compensate for the forced laborer and comfort women led to a trade war between the two countries.67 As a result, the Republic of Korea abruptly declared it would consider not extending the GSOMIA.68 This shocked both Japan and the United States, and it can be observed from how Secretary of Defense Mark Esper and his assistant David Stilwell rushed to South Korea to request the country for a withdrawal from the decision.69 Eventually, after a long negotiation, South Korea did hold on to its final decision of not extending the GSOMIA. This event clearly shows the U.S. that the fate of the trilateral cooperation does not lie on its hands, but rather on the other side of the triangle. Korea – Japan relations are the key for the U.S. to accomplish its plan of ‘trilateral cooperation,’ as Randall Schriver, the assistant Secretary of Defense for Indo-Pacific Security Affairs explains in the speech delivered in August 29th, 2019.70 In the speech, which took place in an event at Center for Strategic and International Studies(CSIS), Mr. Schriver calls both Japan and South Korea as the “cornerstone,” and the “linchpin” in promoting security and peace in the Indo-Pacific region, but also focuses on how the “Japan and Korea feud” are the competitors of the United States in doing so. Without a stable relation between the two, trilateral cooperation might simply be a wish for the United States. Although the tensions between Japan and South Korea have lessened since the beginning of 2019, it is more like an active volcano; no one can expect which tension will drive the two into another big clash. Therefore, trilateral cooperation is not always guaranteed for the United States. V. CONCLUSION The goal of the United States to strengthen military ties with South Korea and Japan against China does not seem to be an easy task. Difficulties arise not just from the political stance of each administration in Korea and Japan but also from the U.S. strategies or approaches it takes to the two Northeast Asian allies. For instance, the U.S. requests to Korea and to Japan to accept a radical increase in shared defense-costs will not help achieve the U.S. goal. Moreover, the aloof or indifferent attitude of the United States towards history issues between Korea and Japan will aggravate the situation. The United States also needs to provide Korea with alternative economic incentives if the Trump administration wants Korea to consider the U.S. request. Since the Korean economy is heavily dependent on China, Korea finds it extremely hard to make a decision that will have serious economic implications. The three suggestions from the United States to Japan and Korea seem therefore to be impractical or unattractive. South Korea and Japan are currently taking steps what they think best, so this paper suggests that the United States find more effective incentives or strategies to bring South Korea and Japan closer to its side. Without any substantial changes, the United States will continue its lonely fight.

66 Masaru, Satō. “The Key US Role in Keeping GSOMIA Alive.” Nippon.com, 30 May 2020, www.nippon.com/en/news/l00255/the-key-usrole-in-keeping-gsomia-alive.html; Cha, Victor. “The Meaning of GSOMIA Termination: Escalation of the Japan-Korea Dispute.” Center for Strategic and International Studies, 13 Aug. 2020, www.csis.org/analysis/meaning-gsomia-termination-escalation-japan-korea-dispute. 67

Sang-hun, Choe. “South Korean Court Orders Mitsubishi of Japan to Pay for Forced Wartime Labor.” The New York Times, The New York Times, 29 Nov. 2018, www.nytimes.com/2018/11/29/world/asia/south-korea-wartime-compensation-japan.html.

68,69

Masaru, Satō. “The Key US Role in Keeping GSOMIA Alive.” Nippon.com, 30 May 2020, www.nippon.com/en/news/l00255/the-key-usrole-in-keeping-gsomia-alive.html. 70 Cha, Victor, “The Importance of U.S.-Japan-Korea Trilateral Defense Cooperation.” Center for Strategic and International Studies. 28 Aug. 2019. https://csis-website-prod.s3.amazonaws.com/s3fs-public/publication/190828_importance_us_japan_korea.pdf

BIBLIOGRAPHY Allen-Ebrahimian, Bethany. “O'Brien Says China's Global Influence Puts Americans at Risk.” Axios, 24 June 2020, www.axios.com/robert-obrien-china-influence-americans-risk-c01a3998-be33-4d8f-9e6c-b60a9185a904.html. Cha, Victor, “The Importance of U.S.-Japan-Korea Trilateral Defense Cooperation.” Center for Strategic and International Studies. 28 Aug. 2019. https://csis-website-prod.s3.amazonaws.com/s3fspublic/publication/190828_importance_us_japan_korea.pdf Cha, Victor. “The Meaning of GSOMIA Termination: Escalation of the Japan-Korea Dispute.” Center for Strategic and International Studies, 13 Aug. 2020, www.csis.org/analysis/meaning-gsomia-termination-escalation-japan-korea-dispute. Sang-hun, Choe. “South Korean Villagers Protest Plans for U.S. Missile Defense System.” The New York Times, The New York Times, 13 July 2016, www.nytimes.com/2016/07/14/world/asia/south-korea-thaad-us.html. Sang-hun, Choe. “South Korean Court Orders Mitsubishi of Japan to Pay for Forced Wartime Labor.” The New York Times, The New York Times, 29 Nov. 2018, www.nytimes.com/2018/11/29/world/asia/south-korea-wartime-compensation-japan.html. Choi, He-suk. “Seoul and Tokyo Dig in over Trade, GSOMIA.” The Korea Herald, 17 Nov. 2019, www.koreaherald.com/view.php?ud=20191117000135. Council on Foreign Relations. “Timeline: U.S. Relations With China 1949–2020.” Council on Foreign Relations, Council on Foreign Relations, www.cfr.org/timeline/us-relations-china. Delaney, Robert. “Pompeo Urges China's Citizens to Help Change Beijing's 'Behaviour'.” South China Morning Post, 24 July 2020, www.scmp.com/news/china/diplomacy/article/3094467/us-secretary-state-urges-chinas-citizens-help-changebehaviour. Erickson, Andrew S. “U.S.-China Military-to-Military Relations: Policy Considerations in a Changing Environment.” Asia Policy, vol. 26, no. 3, July 2019, pp. 123–144., doi:10.1353/asp.2019.0041. Gady, Franz-Stefan. “Toothless Tiger: Japan Self-Defence Forces.” BBC News, BBC, 15 Oct. 2015, www.bbc.com/news/worldasia-34485966. Global Firepower, “Japan Military Strength (2020).” Global Firepower - World Military Strength, www.globalfirepower.com/country-military-strength-detail.asp?country_id=japan. Glosserman, Brad. “Seoul Draws Wrong THAAD Lessons.” The Japan Times, 27 Jan. 2020, www.japantimes.co.jp/opinion/2020/01/27/commentary/japan-commentary/seoul-draws-wrong-thaad-lessons/. Green, Michael J. “The Legacy of Obama's ‘Pivot’ to Asia.” Foreign Policy, Foreign Policy, 29 May 2019, foreignpolicy.com/2016/09/03/the-legacy-of-obamas-pivot-to-asia/. Journal, The Asia Pacific. “Moon Jae-In's THAAD Conundrum: South Korea's ‘Candlelight President’ Faces Strong Citizen Opposition on Missile Defense.” The Asia-Pacific Journal: Japan Focus, 15 Sept. 2017, apjjf.org/2017/18/Martin.html. Lee, Hae-Young, 일, 미방위분담금 5 배 증액 요구에 ‘곤혹’…”불가능에 가깝다” [‘Frustration’ of Japan with 5 fold increase in request of defense cost sharing by the U.S. … “Nearly impossible”], Yonhap News, 2019. https://www.yna.co.kr/view/AKR20190801062800073 Lee, Seong-Hyon. “The Shift of Security Environment in Northeast Asia: The US–China Conflict and Its Implications for Korea.” Journal for Peace and Nuclear Disarmament, vol. 1, no. 2, 2018, pp. 352–362., doi:10.1080/25751654.2018.1542655. Masaru, Satō. “The Key US Role in Keeping GSOMIA Alive.” Nippon.com, 30 May 2020, www.nippon.com/en/news/l00255/the-key-us-role-in-keeping-gsomia-alive.html.

Panda, Ankit. “Japan Suspends Aegis Ashore Missile Defense Plans. What Happens Now?” – The Diplomat, For The Diplomat, 23 June 2020, thediplomat.com/2020/06/japan-suspends-aegis-ashore-missile-defense-plans-what-happens-now/. Panda, Ankit. “What Is THAAD, What Does It Do, and Why Is China Mad About It?” – The Diplomat, For The Diplomat, 26 Feb. 2016, thediplomat.com/2016/02/what-is-thaad-what-does-it-do-and-why-is-china-mad-about-it/. Park, Boram. “S. Korean Defense Minister Denies Bolton's Claim on Defense Cost-Sharing Deal.” Yonhap News Agency, 24 June 2020, en.yna.co.kr/view/AEN20200624008700315. Reif, Kingston. “Arms Control Today.” Arms Control Association, Sept. 2017, www.armscontrol.org/act/2017-09/news/moonreverses-thaad-decision. Sieg, Linda. “Mission Unaccomplished - Abe's Drive to Revise Pacifist Constitution.” The Japan Times, 19 Nov. 2019, www.japantimes.co.jp/news/2019/11/19/national/politics-diplomacy/shinzo-abe-revise-constitution/. Shihyeon, Jo. “문재인 대통령 ‘자주국방 힘 위에서 한미동맹 굳건해진다.’” Newbc 뉴비씨, 19 July 2019, www.newbc.kr/news/articleView.html?idxno=6170. Shortell, David. “FBI Director Unleashes on China in Speech.” CNN, Cable News Network, 7 July 2020, edition.cnn.com/2020/07/07/politics/christopher-wray-fbi-director-china-speech/index.html. Stalenheim, Petter. “Military Expenditure” SIPRI Yearbook 2008: Armaments, Disarmament and International Security. Jan. 2019. https://www.sipri.org/sites/default/files/YB08%20175%2005.pdf Stewart, Taylor. “Does Japan Have an Army?: Culture: Metropolis Magazine Japan.” Metropolis Japan, 9 Aug. 2019, metropolisjapan.com/culture-japanese-self-defense-forces/. Stiles, Matt. “Upset over a U.S. Missile Defense System, China Hits South Korea Where It Hurts - in the Wallet.” Los Angeles Times, Los Angeles Times, 28 Feb. 2018, www.latimes.com/world/asia/la-fg-china-south-korea-tourism-20180228htmlstory.html Stracqualursi, Veronica. “Trump Apparently Threatens to Withdraw US Troops from South Korea over Trade.” CNN, Cable News Network, 16 Mar. 2018, edition.cnn.com/2018/03/15/politics/trump-us-troops-south-korea/index.html. Timmons, Heather. “Timeline: Key Dates in the U.S.-China Trade War.” Edited by Leslie Adler, Reuters, Thomson Reuters, 16 Jan. 2020, www.reuters.com/article/us-usa-trade-china-timeline/timeline-key-dates-in-the-u-s-china-trade-waridUSKBN1ZE1AA. United States, Special Measures Agreement, 1991, state.gov, https://www.state.gov/wp-content/uploads/2019/06/19-405-KoreaSMA.pdf United States, Status of Forces Agreement, 1966, usfk.mil, https://www.usfk.mil/Portals/105/Documents/SOFA/A01_SOFA.Art.IXXXI.pdf “US and South Korea Agree THAAD Missile Defence Deployment.” BBC News, BBC, 8 July 2016, www.bbc.com/news/worldasia-36742751. Wicker, Mcdaniel. “Completing the Triangle: Executive Summary.” Wilson Center, 1 Dec. 2019, www.wilsoncenter.org/publication/completing-the-triangle-executive-summary. National Defense Authorization Act 2017 United States, Congress, Cong. Senate, Armed Services. National Defense Authorization Act for Fiscal Year 2017: Conference Report to Accompany S. 2943, U.S. Government Publishing Office, 2016. 114th Congress, 1st session, bill S.2943. 2018 21

United States, Congress, Cong. House, Armed Services. National Defense Authorization Act for Fiscal Year 2018: Report of the Committee on Armed Services, House of Representatives, on H.R. 2810, Together with Additional Views (Including Cost Estimate of the Congressional Budget Office), U.S. Government Publishing Office, 2017. 115th Congress, 1st session, bill H.R.2810. 2019 United States, Congress, Cong. House, House - Armed Services. John S. McCain National Defense Authorization Act for Fiscal Year 2019, U.S. Government Publishing Office, 2018. 115th Congress, 1st session, bill H.R 5515. 2020 United States, Congress, Cong. Senate, Armed Services. National Defense Authorization Act for Fiscal Year 2020: Conference Report to Accompany S. 1790, U.S. Government Publishing Office, 2019. 116th Congress, 1st session, bill S.1790

Defense White Paper of Republic of Korea Republic of Korea, Ministry of National Defense, 2018 Defense White Paper, Ministry of National Defense, 31 Dec. 2018, https://www.mnd.go.kr/user/mnd/upload/pblictn/PBLICTNEBOOK_201907110548253080.pdf

Image Processing of Brain CT in Acute Cerebral Infarction Beomseok Kim

Ripon Christian High School in the U.S. ilovejacobkim@gmail.com

used to reduce noise and extract features of such diagnostic images. Gaussian blur and median filter are both image smoothing techniques [6], which reduce noise. Gaussian blur suppresses the noise and provides edge detection in the image [7]. Median filter arranges pixel values in a vector format, in which central value would be taken in the rank-ordered distribution [8]. Lee et al. has shown that an adaptive partial median filter has improved the diagnosis of acute cerebral infarction using non-enhanced CT images [9].

Abstract— Radiologic tasks nowadays utilize computer-aided diagnosis to interpret and diagnose diseases. For emergency symptoms like acute cerebral infarction, accurate diagnosis from early CT image is crucial, but the ambiguity of findings such as hypoattenuation of image and loss of the gray-white matter interface makes it difficult to discern. The purpose of imaging processing is to aid radiologists in finding discrete lesions of cerebral infarction by enhancing the image and removing ambiguity. Using Python program language with the OpenCV library, filters were created to enhance a set of verified CT acute cerebral infarction images. Gaussian filters and median filters were applied to reduce noise. Histogram equalization was also performed on respective filters to further enhance the image. Among the methods, histogram equalization of the median filter has provided substantial results in interpreting the image due to its clarity on borders with less salt-and-pepper noise. Keywords: CT, stroke, acute cerebral infarction, image processing, histogram equalization

Histogram equalization is another method of image processing. By adjusting contrast, the intensity is evenly distributed on the histogram, increasing the overall resolution by using the cumulative distribution function. Histogram equalization is useful in image processing for images of extremely bright, dark, and low-contrast images. Thus, the method is suitable in medical imaging [10], such as MRI or CT, in which the background of the image is usually black and white. In this paper, imaging processing was utilized with gaussian blur filter, median filter, and histogram equalization to determine the lesion of cerebral infarction in a partial image section of CT.

I. INTRODUCTION Image processing and computer vision techniques are being developed to facilitate radiologic tasks through a paradigm known as computer-aided diagnosis (CAD) [1]. Since the 1980s, when the systematic investigation began, CAD has been growing as a dynamic research field to where today radiologists use the computer output as a “second opinion” to make final decisions [2]. A CAD system aims to help improve diagnostic accuracy for radiologists and may consist of stages such as noise reduction, smoothing, feature extraction, and classification [3]. These image processing stages are used in various diagnoses of diseases such as breast cancer, lung cancer, cardiovascular diseases, and colon cancer. Among the organs of application is the brain, where medical modalities as CT and MRI are commonly used.

II. METHODS A set of CT images of an 82-year-old-female with acute cerebral infarction in the middle cerebral artery (MCA) territory on the left hemisphere (Fig.1a) was prepared with a corresponding MRI image (Fig.1b). MRI image was referenced to verify the lesion of acute cerebral infarction.

Acute cerebral infarction is one of the major cerebrovascular diseases, and the detection of its early signs is crucial for one’s survival. CT is still the most commonly used imaging modality for the diagnosis of acute stroke because of its wide availability and examination of speed and price. Yet, the advanced MRI is superior to non-enhanced CT in respect of sensitivity [4]. Early CT signs such as the loss of gray-white matter interface are subtle to detect [5]. To compensate, image processing techniques could be used in CT analysis to improve its diagnostic accuracy, especially when it comes to urgent situations such as stroke.

Figure 1. a. CT of left cerebral infarction. b. Diffusion MRI of left cerebral.

Image processing was done with OpenCV library and programming language Python. Two sections were conducted for image processing. First, image filtering methods of gaussian blur and median filter were used. Grayscale, or contrast preserving decolorization, was performed to the CT image to store the intensity of light through the black and white spectrum. Then, gaussian blur and median filter were applied to the grayscale image of CT.

Image processing techniques include gaussian blur filter, median filter, and histogram equalization. Image filters are

For the second section, the image was processed with histogram equalization, increasing contrast and distributing intensity. The histogram equalization was taken on grayscale, gaussian blur, and median filter.

Beomseok Kim is with the Class of 2021, Ripon Christian High School, Ripon, CA, 95366, USA (e-mail: ilovejacobkim@gmail.com)

in the deficiency of filtering methods in detecting cerebral infarction.

III. RESULTS Image processing techniques of filtering methods such as gaussian blur and median filter were used to reduce noise and contrast in the image (Fig. 2). Compared to grayscale, gaussian blur and median filter increased the smoothness of an image. Median filter had a high resolution of pixels on edge compared to that of gaussian blur. Both median filter and gaussian had the kernel size of 7 to compare the extent of enhancement.

The second approach of using histogram equalization enabled the acute cerebral infarction distinguishable. While preserving image resolution, contrast and brightness were drastically enhanced. However, histogram equalization had a high contrast image consisting of salt-and-pepper noises [12]. As some subtle loss of gray-white matter interfaces in acute cerebral infarction appeared due to image noise [5], it is also essential to reduce noise when increasing image contrast. The solution for these conditions was taking the histogram equalization of the median filter. This produced an image with less noise, yet it also provided high contrast to enable visual detection of acute cerebral infarction in the left hemisphere.

Figure 2. CT image with filters applied: Respectively Grayscale (Original), Gaussian Blur, and Median Filter

Histogram equalization was used to enhance the discernment of cerebral infarction (Fig. 3). Overall, this method decreases the ambiguity of the gray matter border, which shows similar image from that of diffusion MRI (Fig.1b).

REFERENCES [1]

S. G. Armato III et al., “Lung Image Database Consortium: Developing a Resource for the Medical Imaging Research Community,” Radiology, vol. 232, no. 3, pp. 739–748, Sep. 2004, doi:10.1148/radiol.2323032035. [2] K. Doi, “Computer-aided diagnosis in medical imaging: Historical review, current status and future potential,” Comput. Med. Imaging Graph., vol. 31, no. 4–5, pp. 198–211, Jun. 2007, doi: 10.1016/j.compmedimag.2007.02.002. [3] M. K. Abd-Ellah, A. I. Awad, A. A. M. Khalaf, and H. F. A. Hamed, “Design and implementation of a computer-aided diagnosis system for brain tumor classification,” in 2016 28th International Conference on Microelectronics (ICM), Dec. 2016, pp. 73–76, doi: 10.1109/ICM.2016.7847911. [4] W. J. Powers et al., “Guidelines for the Early Management of Patients With Acute Ischemic Stroke: 2019 Update to the 2018 Guidelines for the Early Management of Acute Ischemic Stroke: A Guideline for Healthcare Professionals From the American Heart Association/American Stroke,” Stroke, vol. 50, no. 12, Dec. 2019, doi: 10.1161/STR.0000000000000211. [5] N. Tomura, K. Uemura, A. Inugami, H. Fujita, S. Higano, and F. Shishido, “Early CT finding in cerebral infarction: obscuration of the lentiform nucleus.,” Radiology, vol. 168, no. 2, pp. 463–467, Aug. 1988, doi: 10.1148/radiology.168.2.3393665. [6] J.-S. Lee, “Digital image smoothing and the sigma filter,” Comput. Vision, Graph. Image Process., vol. 24, no. 2, pp. 255–269, Nov. 1983, doi: 10.1016/0734-189X(83)90047-6. [7] G. Deng and L. W. Cahill, “An adaptive Gaussian filter for noise reduction and edge detection,” in 1993 IEEE Conference Record Nuclear Science Symposium and Medical Imaging Conference, pp. 1615–1619, doi: 10.1109/NSSMIC.1993.373563. [8] M. Nixon and A. Aguado, Feature Extraction and Image Processing for Computer Vision 4th Edition. 2019. [9] Y. Lee, N. Takahashi, D.-Y. Tsai, and K. Ishii, “Adaptive partial median filter for early CT signs of acute cerebral infarction,” Int. J. Comput. Assist. Radiol. Surg., vol. 2, no. 2, pp. 105–115, Jul. 2007, doi: 10.1007/s11548-007-0123-3. [10] Yeong-Taeg Kim, “Contrast enhancement using brightness preserving bi-histogram equalization,” IEEE Trans. Consum. Electron., vol. 43, no. 1, pp. 1–8, 1997, doi: 10.1109/30.580378. [11] Wardlaw, J. M., Dorman, P. J., Lewis, S. C., & Sandercock, P. A. G. (1999). Can stroke physicians and neuroradiologists identify signs of early cerebral infarction on CT? Journal of Neurology, Neurosurgery & Psychiatry, 67(5), 651–653. https://doi.org/10.1136/jnnp.67.5.651. [12] Jayaraman, Digital Image Processing. 2009. pp. 272-304

Figure 3. CT image with histogram equalization: Respectively Histogram Equalization of Grayscale, Gaussian Blur, and Median Filter

IV. DISCUSSION Difficulties in determining cerebral infarction have been a major problem in medicine [11]. The image taken by MRI shows clear evidence of left MCA territory infarction. In contrast, the CT image produces ambiguity as the boundaries between the gray and white matter is unclear. Thus, CT images could cover the role of MRI if CT images could show cerebral infarction in the brain with the use of image processing. This paper proposed different methods of grayscale, gaussian blur, median filter, and histogram equalization. The histogram showed similar data for each method, for filters did not significantly change the intensity. Gaussian blur and median filter provided similar results (Fig. 2), which both added smoothness and noise reduction compared to the original image. While the median filter preserved resolution at the edge area, the gaussian blur did not. The histogram for each processed image shows that the image processed with the median filter has more pixels of both pure black and white compared to the gaussian blur, which indicates a stronger contrast. Both filtering methods, however, did not contribute to locating the cerebral infarction. According to Lee et al., gaussian and median filter had a high standard deviation rate but a low edge slope rate compared to other adaptive filters [9]. Another approach was required to fill

A Comparison of Gaussian Filtering and Fourier Transform Techniques for Image Denoising Kyuseung Shin and Fourier Transform A Comparison of Gaussian Filtering Chadwick International in the Republic of Korea Techniques for Image Denoising kshin2021@chadwickschool.org Kyuseung Shin _____________________________________________________________________________________

Abstract: Image denoising has been extensively studied in deep learning over the last several years. However, deep learning is always subject to data bias, which may create inaccurate results especially when the input data is irrelevant to testing data. In order to denoise images without bias, the use of analytical techniques is necessary. However, different analytical techniques have different properties, so comparing their individual characteristics is necessary. In this paper, we present a comparative study of different analytical techniques for image denoising. More specifically, we compare different linear filters that are commonly used for analytical image denoising. We first describe the individual filters and their applications for image denoising. Then, we compare the denoised images with the original images through a quantitative and qualitative analysis by comparing different PSNR values of denoised and original images, and by comparing denoised results. Next, we suggest situations in which certain methods are most effective based on the analysis. Finally, we note some limitations to the study and suggestions for future research. Keywords: Image Denoising, Additive White GaussianNoise, Linear Filters, GaussianFiltering, Fourier Transform, Image Processing, Computer Vision _____________________________________________________________________________________

1. Introduction Digital images are widely used in various science and engineering fields, such as medical imaging and human recognition [1][31]. However, they are often affected by noise that can degrade the information present in the image [2]. Therefore, image denoising has been extensively studied over the past 50 years [3]. While traditional analytical denoising methods are able to output accurate results in most cases, significant drawbacks exist such as the need for manually setting parameters and optimizing the system [4]. When the analytical procedures become more advanced, the complexity of parameter optimization increases [4]. In recent years, deep learning methods for denoising and image processing in general gained momentum [4] to outperform the traditional analytical methods -- the deep learning methods were more flexible, meaning that there was no need to manually calculate and optimize parameters [3]. However, the prominence of deep learning for image denoising does not necessarily mean that traditional analytical methods are obsolete. Indeed, the training process of deep neural networks (DNNs)

relies on datasets of finite size [7]. However, any finite dataset is biased as it cannot include all elements of the world [7]. Therefore, DNNs cannot create good results when the input data is not relevant to the training set, especially when the dataset is heavily biased [7]. For these reasons, there is a need to compare different analytical denoising techniques, which do not utilize DNNs, for an unbiased analysis. In this research, we compare different linear filters that are used for denoising. Linear filters modify the signal through the filter’s transfer function [21]. A convolution of the input image with the linear filter operator creates the output image [21]. The linear filters we compare in this research can be divided into two large categories: Gaussian filtering (using a Gaussian kernel [23]) and filtered Fourier transform. Variants of the Gaussian filter will include weighting the edges of the noisy image found by Sobel [11] and Canny [12] edge detection techniques and combining them to the Gaussian filtered results. Such methods help define clearly the edges found in an image in order to overcome one of the fundamental drawbacks of Gaussian filtering, which is the blurring effect that degrades the edges [10]. To compare these Gaussian denoising methods (all of these methods use Gaussian filtering) for image denoising, we also compare different Fourier transform linear filtering methods (that operate in the frequency domain): Fourier Transform with a combination of low and high pass filters [19], Fourier Transform with Butterworth Filter [19], Fourier Transform with Wiener filter [17], and finally BM3D [18]. While the BM3D algorithm is technically not a type of Fourier transform, it still converts the image into the frequency domain (using discrete cosine transform) over a small window, making the method similar to Fourier transforms [18]. The reason for comparing various filters is the fact that different filters have different capabilities. Varying the types of filters would allow for a more thorough investigation of the differences between applying Gaussian filters and linear filters on Fourier (or cosine, for BM3D) transformed images for image denoising. To ensure consistency, we use synthetic noise. Specifically, white Gaussian noise (AWGN) is added to the grayscale images [3]. Since it is possible to tune the amount of noise applied to a certain image, we are able to make meaningful comparisons of the denoised at comparable relative noise levels. The noise level is varied by tuning the parameter ı while the amplitude will be fixed 1, where the image pixel’s maximum value is 1 and minimum is 0. The noise is added to the 12 images from the set12 dataset (found here: [13][14][15]) that is used to test the quality of each different filter. The set12 dataset contains 12 grayscale images (of either 256x256 px or 512x512 px) that are most commonly utilized for image processing research. These 12 images all have different characteristics (for instance, the house image has minimal detail while the image of lenna has intricate details around her hair, yet other parts of the image is still simplistic), thus processing all 12 through the different methods will decrease the bias that arises due to different capabilities and advantages of each filter. This paper aims to compare two large branches of analytical denoising methods, Gaussian filtering and Fourier transform, by comparing image denoising quality (measured with PSNR [9]). Specifically, we compare PSNR values by computing small size images (64x64 pixels, 128x128 pixels, 256x256 pixels, and 512x512 pixels). Small images will have relatively less features (compared to larger images due to there being less pixel data), therefore denoising results may vary even if they are of the same image. Thus, there is room for comparison between different image sizes as well as the type of images themselves. The paper is ordered as follows. We introduce the specifics of the individual Gaussian and Fourier denoising methods that were used for comparison in sections 2 and 3. The methodology for comparing the outputs of the algorithms is discussed in section 4. The results and the interpretation of the data will be discussed

in section 5. Finally, relevant conclusions on the data and some comments on future studies will be documented in section 6.

2. Overview of Different Gaussian Denoising Methods This section gives a brief overview of how Gaussian filters work, and how the edge-detection methods that are used in together with the Gaussian filter operate. Gaussian filtering utilizes spatial distances between pixels and their different values to smooth out the image and reduce noise [10]. Gaussian filtering uses a kernel (of different sizes, but the sum of the matrix values should equate to 1) to take into consideration all the pixel values in close proximity. As the noise will most likely not cover a vast area completely, it will be eliminated by the influence of close pixels that are not affected by noise. Obviously, the closer the pixel is to the target pixel, the more influence it will have, which is demonstrated by larger numbers in the kernel matrix, as shown in a) of Fig. 1., where the center pixel has a weighting of 25%, the largest out of the 9 pixels.

Fig. 1. a) shows an example of a 3x3 Gaussian Kernel, b) shows an image of the Camera Man (size 256x256 found in set12 dataset, found in sources [13][14][15]) in which additive white Gaussian noise is applied to where =0.0, ı=0.1, c) shows an image of the Camera Man that is denoised using the Gaussian kernel. In the third image, it is possible to recognize that the person’s edges are blurred in c) compared to b), a result of smoothing the entire image.

However, Gaussian filtering does have a disadvantage of not preserving edges completely [10]. As the kernel is slid over an edge, the pixels will have high contrast (as there is abrupt color change on the edges) that will be smoothed out. As shown in Fig.1., the filtered image’s edge is slightly blurred out and the camera stand also lost some of the reflections that characterized detail. 2.1. Sobel Filters In order to resolve the issues of Gaussian filtering (blurring out edges [10]), edge detection methods are used that add the edges to the filtered image. One common edge-detection method is the Sobel filter [11]. The Sobel filter has two kernels: one responsible for detecting edges horizontally, and another for detecting edges vertically [20]. A particular feature of the kernel is that the kernel is designed to complement pixel values opposite but of equal displacement of each other when they are of the same value [20]. For instance, if the direct left and right pixels of a particular pixel has the same value, then the kernel calculation negates both of those values by subtracting them. However, on the other hand, if the left and right pixels have opposite values (such as a white left pixel and black right pixel), then the values adds up, showing that there is indeed an edge. Due to the nature of the design, one Sobel kernel can only obtain edges in one direction [20]: it means that for a vertical kernel, vertical edges will be better detected compared to horizontal edges, and vice versa. Therefore, the edges found from the two kernels of

different directions (Sobel kernels can have maximum of 4 directions [20], but the general direction can be simplified to horizontal and vertical) are combined together in the following method: Ef =

¥(E ) h

+ (E v )2

where Ef denotes the final edge matrix, and Eh and Ev denotes the horizontal and vertical edge matrices respectively. Afterwards, the edges are combined with the Gaussian filtered image in order to more clearly define the edges that are smoothed out by the Gaussian kernel. However, in this process, the pixel value of the edges should be decreased as well. In this particular research, it was found that multiplying 0.01 to Ef produced the best denoising results. 2.2. Canny Edges Another common edge detection technique is the Canny algorithm [12]. The Canny edge detection method is a multi-step process that utilizes a filter based on the derivative of a Gaussian [22]. This filter computes the intensity of the gradients, after which only the highest intensity is selected to create a one pixel thick edge. At this stage, it is possible for noise to remain as pixels with noise will also have contrasting gradient values [22]. Therefore, the intensity of the gradient is again filtered to remove certain ones below the threshold, thus effectively removing the noise [22]. In this process, there is also a high-pass threshold in which “strong edges” are categorized. Any other preserved edge pixels are classified as “weak edges”. As some of the weak edges may result from image noise, it is blob-checked to see if there are any immediate links to a strong edge pixel [22]. In this case, the weak edge pixel is preserved, while otherwise it is discarded.

3. Overview of Different Fourier Transform Denoising Methods This section will give a brief overview of how different Fourier transform filtering algorithms work. Fourier transform by itself is a mathematical theory concerned with dividing a particular function into its constituent frequencies [16]. When Fourier transforms are utilized for image denoising, it first translates the image into its frequency components [16]. When the frequency domain is created, smaller coefficients represent low frequency and the larger coefficients represent high frequency [6]. Image noises are included in that high frequency domains, so a filter to mask the frequency domain and then inverse Fourier transforming it to form the image will be able to yield a denoised image. After the image is converted into the frequency domain, it is important for the domain to be processed by the right type of linear filter that is able to clearly distinguish between noise and non-noise components of the image. Below, the specific explanations for the filters and methods used in this particular research, low and high pass filters, Butterworth filters, Wiener filters, and finally, BM3D, will be given [19][17][18]. 3.1. Low and High Pass Filters for Fourier Transform A typical denoising method is to combine low and high pass filters to mask the frequency domain produced by Fourier transform. A low pass filter (LPF) is a filter that aims to only pass frequencies lower than the selected cutoff frequency [25]. In Fourier transform, as areas of high frequency typically consist of noise and edges, passing only the low frequency areas achieves an effect similar to denoising.

However, this also leads to a side effect of the edges blurring, thus degrading the original quality of the image unintentionally. On the other hand, a high pass filter (HPF) is a filter that passes frequencies higher than the selected cutoff frequency [26]. The high pass filter has the capability to preserve edges [27], which is why it can be used in parallel with the low pass filter to create a denoised image. Thus, once the selected areas of low and high frequencies are conserved, the frequency domain is inverse Fourier transformed to form the denoised image. 3.2. Butterworth Filters for Fourier Transform The Butterworth filter acts similarly to the low and high pass filters that are described above [28]. However, the Butterworth filter has an advantage in that it can smooth out the ripples created from some other filters [29], which contributes to increasing the accuracy of the filter. The Butterworth filter can be created in numerous orders of n, which allows the program to change the sharpness of the filter based on the order [28]. For this research, the order of the butterworth filter was set to n=1. The generic versions of the Butterworth low and high pass filters are designed by the following [28]: LPF: H [u, v ] = 1/(1 + (D[u, v ]/D0 )2 )

HPF: H [u, v ] = 1/(1 + (D0 /D[u, v ])2 ) where D0 is the cutoff frequency. For the low pass filter, as the cutoff frequency increases, the output image becomes less blurred. On the other hand, the increase in cutoff frequency for the high pass filter creates more refined and sharp edges that the low pass filter wasn’t able to preserve [28]. Overall, the Butterworth filter acts similar to the low and high pass filter depicted in 3.1, but is able to, in theory, create better results due to the ability to smooth out the ripples in the frequency domain [29]. 3.3. Wiener Filters for Fourier Transform The Wiener filter is a type of filter that attempts to produce an estimate of a signal from a corrupted signal, usually due to noise [24]. Therefore, the frequency domain created by the Fourier transform operation is then filtered by the Wiener filter to create an estimate signal [24]. Afterwards, the output domain is inverse Fourier transformed into the original image. The specific filter takes into account the correlation between the noise of the signal and the noise-corrupted signal to produce an estimate, and attempts to decrease the mean square error as much as possible [32]. To take the elements into account, the Wiener filter estimates the ratio of power spectrum of the noise to that of the original image [5]. For this research, a 5x5 kernel is Fourier transformed into the frequency domain in order to construct the Wiener filter. 3.4. BM3D As expressed before, the BM3D algorithm is not technically a variant of the Fourier transform denoising method. However, in a broad sense, both Fourier transform image denoising methods and BM3D [18] all operate in the frequency domain [8] and process the images by applying linear filters to frequency values, thus showing relative similarity. The BM3D algorithm is designed with two parts: BM stands for block matching, and 3D stands for 3D transform for collaborative filtering. Before the main

stages of the algorithm, the image is first converted through discrete cosine transform (DCT), which is different from Fourier transform in that it only preserves the cosine function. In the initial stages, the image is divided into small sections and compared to each other for similarity in pixels [18]. This is the block matching process in which the separate sections of the image denoted as “blocks” are categorized with each other. The principle is that in those similar blocks, some of the different elements that the blocks have will have relatively higher chances to be noise. This is because the similar blocks have similar features, so distinctive elements are likely to be noise [18]. After the blocks are grouped and sorted, the individual sections on the blocks are collaboratively filtered. It is assumed that without noise, all the blocks are identical (this assumption will not be applicable for most cases in real life, but the blocks are to be similar enough so that mean estimation will not cause significant quality degradation). Afterwards, the blocks are aggregated with each other then goes through inverse DCT to recreate the first estimate of the noise removal process [18]. The second (and final) estimate of the noise removal process is similar to the first estimate, as in that the block matching process is done and that these matched blocks are 3D filtered. One minor difference is that the threshold for the block matching process is much higher in the second estimate compared to the first. However, the major difference is that during the 3d filtering process, the filtered results are not applied to the output from the first estimate, but the original image to refine the denoised output [18].

Fig. 2. a) shows an example of the BM3D reconstruction of the cameraman image from the first estimate. It is clear that the image is blurred out heavily and that a lot of the detail, alongside with the noise, are gone. However, b) shows an example of the second (final) estimate that calculated the weights from the first estimate and then applied to the original image. Now, while there are some artifacts, the final image has a lot more detail than the first one.

Thus, BM3D may not seem like a traditional linear filter, however, its collaborative filtering techniques use a combination of signals of individual pixels to process each other, so it does indeed have concepts similar to that of the linear filter.

4. Methodology As briefly depicted above, the image denoising methods are compared primarily by their PSNR (Peak Signal-to-Noise Ratio) values. The PSNR metric is calculated by calculating the root means square error (RMSE) and log scaling it. More specifically, with a reference image f and test image g, the PSNR equation is defined as follows [9]:

PSNR (f , g ) = 10 · log 10 ( 2552 / MSE (f , g ) ) and the mean square error (MSE) is defined as [9]: MSE (f , g ) =

1 MN

M N

(f ij í g ij )2

i=1 j=1

For this research in particular, the reference image f is the original image before noise is added, and the test image g is the denoised image processed by different algorithms. As shown in the PSNR equation, as the mean square error approaches 0 the PSNR value approaches infinity, therefore higher PSNR values indicate higher similarities between images while a low value indicates high differences between the two images [9]. Also, as mentioned before, the dataset we use is set12, and includes some of the most iconically used images in the field of image processing: the cameraman, lenna, monarch, etc. These images have distinct qualities that provide different challenges for the denoising algorithms to overcome. For instance, the lenna image has the hat with feathers which has relatively more complexity in terms of pixel arrangement compared to other parts of the image. Furthermore, the images will be processed in four different sizes (512x512 px, 256x256 px, 128x128 px, 64x64 px). Not all images provided by set12 are of size 512x512 pixels; some are 256x256 pixels. Therefore, while it is possible to resize images to smaller sizes, increasing the image size may lead to decrease in resolution. Thus, when comparing images of size 512x512 px, only 7 images (that were found in set12 and its color version) are used, while the other sizes have all 12 image variants.

5. Results 5.1. PSNR Results on Running Different Methods of Denoising on Varying Image Size and Noise Levels Table 1: PSNR (dB) of Different Methods on Set12 Images of 64x64 px on Different Noise Levels (15, 25, 50) image

gaussian

sobel

canny

low/high

butterworth wiener

noise level

bm3d

avg

best

ᶥ = 15

airforce

30.168

30.175

30.184

29.107

29.783

29.303

29.820

29.791 canny

barbara

29.246

29.232

29.116

29.037

29.116

28.249

28.671

28.952 gaussian

boat

29.584

29.567

29.275

29.122

29.565

28.636

29.033

29.255 gaussian

cman

29.948

30.128

29.631

28.934

29.401

29.291

30.171

29.643 bm3d

couple

29.549

29.595

29.346

29.222

29.475

28.489

28.894

29.224 sobel

house

30.817

30.855

30.473

29.305

30.133

29.909

31.596

30.441 bm3d

lenna

29.981

30.036

29.760

29.194

29.438

28.512

29.597

29.503 sobel

man

29.471

29.490

29.356

29.116

29.509

28.331

28.707

29.140 butterworth

monarch

29.497

29.275

29.082

29.033

28.422

28.877

29.098 gaussian

parrot

29.630

29.653

29.435

28.879

29.412

28.719

28.953

29.241 sobel

peppers

29.776

29.896

29.521

28.984

29.456

28.510

28.997

29.306 sobel

starfish

29.207

29.201

29.185

28.894

29.061

28.219

28.501

28.895 gaussian

noise level

ᶥ = 25

airforce

28.950

28.934

28.937

28.276

28.793

28.552

28.695

28.734 gaussian

barbara

28.830

28.737

28.689

28.415

28.656

28.111

28.236

28.525 gaussian

boat

29.140

29.255

28.955

28.430

28.963

28.473

29.122

28.906 sobel

cman

29.171

29.073

28.707

28.420

28.807

29.006

30.170

29.050 bm3d

couple

29.088

29.040

28.874

28.493

28.943

28.380

28.884

28.815 gaussian

house

29.783

29.732

29.365

28.616

29.305

29.260

31.258

29.617 bm3d

lenna

29.231

29.276

29.012

28.459

28.932

28.512

29.458

28.983 bm3d

man

28.974

29.010

28.885

28.441

28.788

28.266

28.555

28.703 sobel

monarch

28.845

28.823

28.709

28.438

28.653

28.387

28.697

28.650 gaussian

parrot

28.877

28.818

28.638

28.343

28.625

28.397

28.959

28.665 bm3d

peppers

29.196

29.152

28.941

28.423

28.842

28.448

29.102

28.874 gaussian

starfish

28.780

28.721

28.727

28.377

28.634

28.070

28.364

28.525 gaussian

noise level

ᶥ = 50

airforce

27.881

27.990

28.047

28.004

27.948

27.907

27.595

27.910 canny

barbara

28.145

28.142

28.160

28.015

28.205

28.042

28.038

28.107 butterworth

boat

28.325

28.273

28.227

28.039

28.229

28.271

28.784

28.307 bm3d

cman

28.378

28.344

28.278

27.926

28.271

28.286

28.492

28.282 bm3d

couple

28.401

28.369

28.330

27.999

28.271

28.241

28.534

28.307 bm3d

house

28.337

28.312

28.293

27.982

28.315

28.188

28.682

28.301 bm3d

lenna

28.191

28.205

28.189

27.921

28.218

27.977

28.469

28.167 bm3d

man

28.168

28.192

28.139

27.969

28.223

28.075

28.067

28.119 butterworth

monarch

28.125

28.052

28.090

27.991

28.099

28.003

27.995

28.051 gaussian

parrot

28.304

28.227

28.256

27.932

28.164

28.219

28.352

28.208 bm3d

peppers

28.247

28.243

28.211

28.011

28.213

28.102

28.068

28.156 gaussian

starfish

28.068

28.036

28.106

27.955

28.095

27.960

28.095

28.045 canny

Table 2: PSNR (dB) of Different Methods on Set12 Images of 128x128 px on Different Noise Levels (15, 25, 50) image

gaussian

sobel

canny

low/high

butterworth wiener

noise level

bm3d

avg

best

ᶥ = 15

airforce

30.686

30.744

30.566

29.263

30.103

29.904

30.805

30.296 bm3d

barbara

30.071

30.058

29.739

29.255

29.766

28.630

29.532

29.579 gaussian

boat

30.233

30.224

29.843

29.234

29.962

28.954

29.660

29.730 gaussian

cman

30.351

30.395

30.003

29.048

29.862

29.784

30.556

29.999 bm3d

couple

29.947

29.988

29.599

29.247

29.821

28.720

29.496

29.546 sobel

house

31.642

31.589

30.906

29.453

30.613

30.661

33.236

31.157 bm3d

lenna

30.938

30.934

30.344

29.378

30.255

29.262

30.962

30.296 bm3d

man

30.280

29.862

29.311

30.007

28.871

29.616

29.747 sobel

monarch

30.545

30.591

29.979

29.309

29.778

29.014

30.214

29.918 sobel

parrot

30.346

30.341

29.901

29.014

29.800

29.214

29.718

29.763 gaussian

peppers

30.768

30.781

30.118

29.212

30.073

29.067

30.424

30.063 sobel

starfish

30.255

30.301

30.031

29.181

29.808

28.550

29.162

29.612 sobel

noise level

ᶥ = 25

airforce

29.212

29.224

29.126

28.393

28.992

28.846

28.969

28.966 sobel

barbara

29.192

29.164

29.029

28.431

29.002

28.465

28.951

28.891 gaussian

boat

29.278

29.271

28.959

28.481

29.069

28.730

29.438

29.032 bm3d

cman

29.310

29.320

28.880

28.409

28.884

29.231

30.431

29.209 bm3d

couple

29.295

29.305

29.079

28.506

29.058

28.602

29.254

29.014 sobel

house

29.985

29.935

29.407

28.513

29.415

29.656

32.434

29.907 bm3d

lenna

29.601

29.643

29.269

28.550

29.224

28.845

30.324

29.352 bm3d

man

29.366

29.345

29.095

28.487

29.126

28.594

29.117

29.019 gaussian

monarch

29.459

29.431

29.148

28.504

29.058

28.711

29.871

29.168 bm3d

parrot

29.092

29.061

28.834

28.326

28.836

28.648

29.349

28.878 bm3d

peppers

29.520

29.571

29.145

28.531

29.166

28.834

30.128

29.271 bm3d

starfish

29.187

29.186

29.044

28.453

28.914

28.316

28.776

28.840 gaussian

noise level

ᶥ = 50

airforce

27.952

27.966

28.071

27.964

28.054

27.892

27.677

27.939 canny

barbara

28.211

28.250

28.187

27.972

28.212

28.111

28.236

28.168 sobel

boat

28.356

28.390

28.310

27.987

28.285

28.276

28.768

28.339 bm3d

cman

28.351

28.423

28.324

27.977

28.240

28.419

28.776

28.359 bm3d

couple

28.378

28.411

28.342

28.019

28.319

28.311

28.795

28.368 bm3d

house

28.362

28.394

28.358

28.056

28.297

28.372

29.010

28.407 bm3d

lenna

28.343

28.347

28.272

28.023

28.276

28.233

28.751

28.321 bm3d

man

28.274

28.266

28.239

28.002

28.253

28.106

28.223

28.195 gaussian

monarch

28.289

28.198

28.218

27.956

28.201

28.115

28.237

28.173 gaussian

parrot

28.264

28.280

28.212

27.975

28.200

28.234

28.534

28.243 bm3d

peppers

28.342

28.313

28.343

28.020

28.272

28.227

28.349

28.266 bm3d

starfish

28.236

28.202

28.195

27.994

28.218

28.018

28.213

28.154 gaussian

Table 3: PSNR (dB) of Different Methods on Set12 Images of 256x256 px on Different Noise Levels (15, 25, 50) image

gaussian

sobel

canny

low/high

butterworth wiener

noise level

bm3d

avg

best

ᶥ = 15

airforce

31.389

31.410

31.212

29.406

30.560

30.290

31.922

30.884 bm3d

barbara

30.761

30.750

30.131

29.260

30.367

29.234

30.712

30.174 gaussian

boat

31.003

31.006

30.342

29.365

30.405

29.466

30.643

30.318 sobel

cman

30.409

30.408

30.059

29.086

29.892

29.796

30.727

30.054 bm3d

couple

30.800

30.808

30.145

29.384

30.340

29.161

30.227

30.124 sobel

house

31.993

32.000

31.069

29.444

30.967

31.066

33.516

31.436 bm3d

lenna

31.838

31.820

30.708

29.438

30.843

29.939

32.434

31.013 bm3d

man

31.215

31.240

30.438

29.420

30.576

29.428

30.576

30.413 sobel

monarch

31.513

31.515

30.546

29.393

30.489

29.895

31.898

30.750 bm3d

parrot

30.499

30.501

30.021

29.018

29.959

29.533

30.228

29.966 sobel

peppers

31.578

31.586

30.573

29.307

30.629

29.857

31.681

30.744 bm3d

starfish

31.011

31.042

30.386

29.244

30.240

28.955

30.144

30.146 sobel

noise level

ᶥ = 25

airforce

29.468

29.539

29.345

28.443

29.161

29.013

29.655

29.232 bm3d

barbara

29.465

29.442

29.159

28.481

29.212

28.791

29.735

29.184 bm3d

boat

29.653

29.657

29.222

28.523

29.264

29.006

30.110

29.348 bm3d

cman

29.272

29.291

28.871

28.401

28.931

29.233

30.532

29.219 bm3d

couple

29.659

29.645

29.277

28.501

29.286

28.882

29.960

29.316 bm3d

house

30.043

30.040

29.438

28.522

29.446

29.897

32.638

30.003 bm3d

lenna

29.960

29.956

29.441

28.497

29.436

29.313

31.425

29.718 bm3d

man

29.734

29.708

29.341

28.504

29.355

28.943

29.821

29.344 bm3d

monarch

29.816

29.784

29.387

28.497

29.281

29.223

30.963

29.565 bm3d

parrot

29.150

29.144

28.871

28.337

28.850

28.857

29.709

28.988 bm3d

peppers

29.841

29.836

29.357

28.486

29.319

29.303

31.127

29.610 bm3d

starfish

29.375

29.378

29.147

28.425

29.046

28.489

29.207

29.010 sobel

noise level

ᶥ = 50

airforce

27.964

28.02

28.076

27.963

28.087

27.894

27.723

27.961 butterworth

barbara

28.263

28.284

28.236

27.988

28.218

28.147

28.484

28.231 bm3d

boat

28.421

28.446

28.370

27.972

28.295

28.384

28.957

28.406 bm3d

cman

28.369

28.391

28.323

27.971

28.189

28.390

28.655

28.327 bm3d

couple

28.471

28.429

28.398

27.988

28.340

28.388

29.010

28.433 bm3d

house

28.438

28.452

28.409

27.986

28.338

28.399

28.883

28.415 bm3d

lenna

28.409

28.434

28.345

28.011

28.319

28.367

29.140

28.432 bm3d

man

28.335

28.329

28.268

27.998

28.249

28.229

28.506

28.274 bm3d

monarch

28.281

28.282

28.242

27.996

28.250

28.193

28.452

28.242 bm3d

parrot

28.262

28.252

28.239

27.967

28.156

28.275

28.847

28.286 bm3d

peppers

28.392

28.398

28.342

28.027

28.269

28.332

28.603

28.338 bm3d

starfish

28.236

28.211

28.159

27.947

28.176

28.081

28.276

28.155 bm3d

Table 4: PSNR (dB) of Different Methods on Set12 Images of 512x512 px on Different Noise Levels (15, 25, 50) image

gaussian

sobel

canny

low/high

butterworth wiener

noise level

bm3d

avg

best

ᶥ = 15

airforce

32.076

32.087

31.819

29.427

30.937

30.825

32.956

31.447 bm3d

barbara

30.871

30.876

30.139

29.382

30.336

29.513

31.514

30.376 bm3d

boat

31.308

31.300

30.544

29.350

30.691

29.690

31.001

30.555 gaussian

couple

31.195

31.211

30.383

29.346

30.554

29.511

30.837

30.434 sobel

lenna

32.217

32.216

31.039

29.434

31.111

30.515

33.238

31.396 bm3d

man

31.720

31.712

30.675

29.434

30.870

29.919

31.410

30.820 gaussian

peppers

31.609

31.612

30.626

29.298

30.829

30.373

32.319

30.952 bm3d

noise level

ᶥ = 25

airforce

29.707

29.750

29.497

28.446

29.288

29.256

30.095

29.434 bm3d

barbara

29.440

29.413

29.104

28.474

29.159

28.879

30.188

29.237 bm3d

boat

29.750

29.739

29.258

28.485

29.330

29.156

30.358

29.439 bm3d

couple

29.778

29.787

29.334

28.487

29.326

29.087

30.466

29.467 bm3d

lenna

30.091

30.086

29.502

28.508

29.512

29.628

32.037

29.909 bm3d

man

29.903

29.883

29.429

28.504

29.449

29.240

30.434

29.549 bm3d

peppers

29.863

29.852

29.338

28.481

29.368

29.587

31.547

29.719 bm3d

noise level

ᶥ = 50

airforce

27.998

28.025

28.097

27.982

28.086

27.925

27.771

27.983 canny

barbara

28.237

28.239

28.196

27.992

28.220

28.157

28.574

28.231 bm3d

boat

28.459

28.476

28.386

27.989

28.298

28.410

29.030

28.436 bm3d

couple

28.481

28.391

27.989

28.311

28.435

29.173

28.466 bm3d

lenna

28.421

28.415

28.345

27.991

28.297

28.410

29.272

28.450 bm3d

man

28.361

28.354

28.290

27.985

28.270

28.293

28.665

28.317 bm3d

peppers

28.401

28.409

28.361

27.986

28.293

28.420

28.875

28.392 bm3d

Table 1, 2, 3, 4 shows different PSNR values for each individual method by the varying amount of noise and image size. Also, the average PSNR values per image of different size and noise levels is calculated

to demonstrate which image type is filtered the best. Finally, the best method is determined per image of different size and noise levels based on their respective PSNR values. The following are observable: 1) For images of size 64x64 pixels, methods utilizing the Gaussian filter seems to be much more effective compared to the other methods. This is most apparent when the noise levels are either ᶥ = 15 or ᶥ = 25, where the average mean percentage error between the PSNR of the best method utilizing the Gaussian filter and the average PSNR for the images is 1.38% and 0.87% respectively. Furthermore, when testing in those noise levels, the methods utilizing the Gaussian filter perform the best roughly 71% of the time in terms of PSNR value. However, when ᶥ = 50, there is a lot more variation: the best PSNR values for 50% of the images are created from either BM3D or the Butterworth filter, while the other 50% are from Gaussian filter and Sobel and Canny additions of the Gaussian filter. 2) For images of size 128x128 pixels, the success of denoising based on PSNR comparisons show that both methods utilizing the Gaussian filter and BM3D worked the best. When ᶥ = 15, BM3D works the best for 33% of the images while Gaussian filter variations works the best for 67% of the images. However, for both ᶥ = 25, 50, BM3D works the best for 58% of the images, and Gaussian filter variations works best 42% of the time. 3) For images of size 256x256 pixels, the BM3D method shows the best results for images with higher noise (ᶥ = 25, 50), while also demonstrating good image denoising results for ᶥ = 15 alongside with methods using the Gaussian filter. When ᶥ = 15, BM3D produces the best results for 6 of the images, while Gaussian, Sobel, and Canny methods produce the best results for the remaining 6. However, the difference in PSNR values between BM3D and the best result from the Gaussian, Sobel, and Canny methods aren’t that significant, with a maximum difference of 1.6 dB for the house image, but less than 1.0 dB for all other images. 4) For images of 512x512 pixels, the BM3D method shows the best results for 76% of the image cases on average for all three noise levels. However, it is still worth noting that all methods, except low and high pass filter and Wiener filter, also seem to produce good results when ᶥ = 15 as their respective PSNR values are above 30.0 dB for all images. 5) When comparing the performance of different linear filters applied after the image is transformed to Fourier space (thus excluding BM3D), the Butterworth filter produces the best results 78% of the time, with the Wiener filter and low/high pass filters coming in second or third. It seems that the first order Butterworth filter is sharp enough to eliminate the noise, but not as sharp as in to eliminate most of the detail. 6) Finally, as a comparison between methods using Gaussian filters and methods applying filters to Fourier (or cosine) transformed images, it is clear that for images of small size and/or low noise levels, Gaussian filtering produces good results, however as the image became larger (thus more features) and had more noise, filters applied after Fourier transformation seems to work better. Of those, BM3D seems to work best. A possibility for the BM3D method being an optimal solution for relatively larger images of higher noise levels may be because of the increase in “similar blocks” that can filter out each other to eliminate the random noise.

5.3. Qualitative Analysis The results of the denoising methods are compared with PSNR values. While the PSNR metric attempts to estimate the quality of the reconstructed image, the measured PSNR value may not necessarily correlate with how humans perceive that the image is well constructed. For instance, significant loss of detail in some areas may contribute to drastically lower PSNR values, however humans may not deem it as significant as the computer calculations do as the rest of the general detail is preserved. Therefore, this section presents some results that demonstrate the reconstruction qualities of the denoising methods in an attempt to analyze the aspects of the denoised image that leads to increases or decreases in PSNR values.

Fig. 3. a) is the noisy (noise level ᶥ = 15) 64x64 pixel image of the house from set12. From b) to h), the images are reconstructions from the different denoising methods from higher PSNR values to lower PSNR values (ie: ordered by: BM3D, Sobel, Gaussian, Canny, Butterworth, Wiener, and Low/High Pass)

Fig. 3. shows how the reconstructed images compare to the original and the noisy image of the house in size 64x64 pixel. While according to the PSNR metric the BM3D method is the most accurate, it is observable that the BM3D method also blurred out detail as well. Conversely, methods such as the Butterworth filter are able to preserve more detail (such as the windows), yet there is more residual noise. For Gaussian filtering variants, while their respective PSNR values are lower than that of BM3D, the methods are able to better retain detail despite there being some degree of residual noise. It seems that the BM3D method leads to better PSNR calculations as the house image contained a lot of blobs that didn’t really have a lot of detail. Thus, even though BM3D blurred out the windows, as the rest of the image (like the gray background and singular-colored roof and walls) didn’t have noise, the results seemed better in the perspective of PSNR. From this observation, it is possible to gather that BM3D may remove most of the noise at the cost of detail. Therefore, denoising small images with BM3D may not be as efficient as the details are more condensed into a smaller frame. On the other hand, the Gaussian filtering methods seem to work quite well. While there is indeed some residual noise that is left over, most of the noise is reduced well while still preserving the features.

Fig. 4. a) is the noisy (noise level = 25) 256x256 pixel image of the monarch from set12. From b) to h), the images are reconstructions from the different denoising methods from higher PSNR values to lower PSNR values (ie: ordered by: BM3D, Gaussian, Sobel, Canny, Wiener, Butterworth, and Low/High Pass). The green boxes for c), d), f) show a close up of a portion of the denoised image for the Gaussian, Sobel, and Wiener method respectively.

Fig. 4. shows the reconstruction of a denoised image (noise level: = 25) compared to the original and noisy image of the monarch. Like the house image, the best reconstructed image in terms of PSNR values is produced by BM3D. However, again, the BM3D method blurs out some of the features (most notably the flowers and the sudden contrast of the spots on the monarch) that are preserved in other methods like the Wiener filter. Nevertheless, more features are preserved in this larger image size compared to the house as BM3D is able to at least preserve the general shape of the flower petals. While variants of Gaussian filtering produce better results than the Wiener filter in PSNR values, the qualities didn’t vary significantly and both methods have residual noise. Comparing different Gaussian denoising methods, the Gaussian filter seems to produce better results because the edge detection done by Sobel and Canny edges, which also preserved some extra noise that may lead to an overall (while slight) decrease in the PSNR value. Comparing the Wiener filter method to Gaussian and Sobel denoising methods in the enlarged green image, it is notable that the Wiener filter actually reduces the noise better than the Gaussian and Sobel filter as the noise particles are less apparent. In this comparison of denoising results for a 256x256 pixel image, it is possible to identify that filters applied after the image undergoes Fourier transformation may see an increase in quality as the image sizes increase. However, Gaussian filtering methods have relatively similar results across different image sizes.

Fig. 5. a) is the noisy (noise level = 15) 512x512 pixel image of Lenna in set12. From b) to h), the images are reconstructions from the different denoising methods from higher PSNR values to lower PSNR values (ie: ordered by: BM3D, Gaussian, Sobel, Canny, Butterworth, Wiener, and Low/High Pass). The green boxes for b), c), f) show a close up of a portion of the denoised image for the BM3D, Gaussian, and Butterworth method respectively.

Fig. 5. shows the reconstructed images of Lenna for the different methods compared to the original and noisy image. As the image gets larger, it is quite apparent that overall the denoising methods improved significantly, which is also backed up by the higher average PSNR value. It is also noticeable that unlike other images that have some residual noise, BM3D is able to smooth everything out, yet the details are smoothed out as well. The Lenna image consists of some regions of high detail (ie: the hair), and some regions of abstractness (ie: the ceiling, her shoulders, etc). BM3D is definitely able to reconstruct the ceiling and Lenna’s shoulders with high degree of accuracy, however the details of her hair is lost (as seen in the zoomed in region). On the other hand, variants of Gaussian filtering and the Butterworth filter method are able to produce the details again quite accurately, however still have some residual noise. More specifically, comparing the results from the Gaussian filter and Butterworth filter, there isn’t a high degree of difference between them, but the zoomed in region shows that the Gaussian filter does slightly have some more blur to it compared to the Butterworth filter. From these three comparisons, it is possible to gather the following: 1) The BM3D method is accurate in suppressing noise overall, but it seems that the particular method works best for images without a lot of detail. For instance, restoring the lower-detail regions of Lenna, the Monarch, and the House is more successful, and the qualities of reconstruction in those particular regions were much better than other images. 2) Variants of the Gaussian filtering method in general seem to be more applicable for smaller-sized images. Images of small sizes seem to work better with these methods, while other methods are able to increase the quality of the denoised image as the image sizes increase. 3) In terms of noise level, the image reconstructions from = 50 didn’t really yield any meaningful results. However, a comparison of images of noise = 15, 25 shows that filters applied after

Fourier transformation in general seems to have better performance compared to Gaussian filtering methods, primarily because they have a lot of residual noise that is left over. 5.4. Limitations on Extremely Noisy Images However, it is observed that denoising results for images with additive white Gaussian noise of variance = 50 didn’t actually yield good results both quantitatively and qualitatively. As shown in Table 1, 2, 3, 4, the average PSNR values for denoised images when = 50 was 28.25 dB, which is much lower than when = 15, 25, where the PSNR values are 30.18 dB and 29.22 dB respectively. It seems that such results occurred as the applied noise degraded a lot of the image features and edges.

Fig. 6. Shows a comparison between the original 256x256 px couple image to the noisy image ( = 50) of a), and the produced denoised images from b) to h) ordered from highest PSNR value to the lowest (BM3D, Gaussian, Sobel, Canny, Wiener, Butterworth, and low and high pass filter)

As shown above, the noisy image has most of its features degraded to the extent that it is not really recognizable. Only some of the most distinct elements of the couple image is visible, such as the desk (whose dark wood contrasts with the light floor) and the man standing (who again is wearing dark clothes in contrast to the relatively light background). However, other features are indeed degraded due to the extensively added noise. Therefore, the denoised results also aren’t really accurate in reconstructing the image. One aspect to note though, is that BM3D is still able to get rid of the grainy particles, however almost all of the detail is blurred out. Therefore, it seems that all of the analytical methods assessed in this research have the inherent inability to denoise images with extensive amounts of noise.

6. Conclusion In this paper, we compare some of the analytical approaches to image denoising. First, we outline how each of these methods work, and then implement them to the same set of images to compare these method’s performance on various types of images of different sizes and noise levels.

Through this investigation, it is quite clear that for images of larger sizes, the BM3D method surpasses the other methods in terms of quality, especially for images with low detail. More generally, the increase in image size and noise level will increase the denoising quality of filters applied after the image’s Fourier transformation. However, when it comes down to smaller images, variants of the Gaussian filtering method seems to work equally well, especially for low noise levels. However, there were some limitations to this research as well: 1) The parameters aren’t fully optimized. While parameters are indeed optimized to some degree, it isn’t optimized perfectly, therefore some methods may have had worse results. 2) There are only 4 variations of image sizes. It is observed that between image sizes of 128x128 pixels and 256x256 pixels, there is a need for transitioning from spatial domain methods to frequency domain methods for better quality. However, that distinct barrier is not yet quite clear. 3) Finally, while the PSNR results seem quite valid, some of the results are questionable in the sense that there still is some residual noise left over. This is especially significant for noisy images with noise level = 50. For future research, it seems that a more detailed investigation into image sizes would be essential to determine the more exact boundary of the image size on which the Gaussian filtering methods work well and efficiently compared to filters applied after Fourier transformation. In this research, the increment in image size was large (there was an increase of image size by a factor of 4 for transitioning from 128x128 px to 256x256 px), thus there should be a more detailed investigation with smaller increments.

References [1] Frangioni, John V. "New technologies for human cancer imaging." Journal of clinical oncology 26.24 (2008): 4012. [2] Boyat, Ajay Kumar, and Brijendra Kumar Joshi. "A review paper: noise models in digital image processing." arXiv preprint arXiv:1505.03489 (2015). [3] Tian, Chunwei, et al. "Deep learning on image denoising: An overview." arXiv preprint arXiv:1912.13171 (2019). [4] Lucas, Alice, et al. "Using deep neural networks for inverse problems in imaging: beyond analytical methods." IEEE Signal Processing Magazine 35.1 (2018): 20-36. [5] Chen, Hao, et al. "Image restoration via improved Wiener filter applied to optical sparse aperture systems." Optik 147 (2017): 350-359. [6] Fan, Linwei, et al. "Brief review of image denoising techniques." Visual Computing for Industry, Biomedicine, and Art 2.1 (2019): 7. [7] Tommasi, Tatiana, et al. "A deeper look at dataset bias." Domain adaptation in computer vision applications. Springer, Cham, 2017. 37-55. [8] Navarro, Laurent, and Jérôme Molimard. "Directional Denoising Using Fourier Spectrum Cloning." Fourier TransformsCentury of Digitalization and Increasing Expectations. IntechOpen, 2019. [9] Hore, Alain, and Djemel Ziou. "Image quality metrics: PSNR vs. SSIM." 2010 20th international conference on pattern recognition. IEEE, 2010. [10] Jiang, Li, Yu Han, and Bin Xie. "Edge-Preserving Image Smoothing Via a Total Variation Regularizer and a Nonconvex Regularizer." Procedia Computer Science 154 (2019): 603-609. [11] Singh, Sanjay, et al. "Comprehensive review and comparative analysis of hardware architectures for sobel edge detector." International Scholarly Research Notices 2014 (2014). [12] Biswas, Ranita, and Jaya Sil. "An improved canny edge detection algorithm based on type-2 fuzzy sets." Procedia Technology 4 (2012): 820-824. [13] Petitcolas, Fabien. Public-Domain Test Images for Homeworks and Projects, University of Wisconsin, homepages.cae.wisc.edu/~ece533/images/.

[14] C/Python/Shell Programming and Image/Video Processing/Compression, www.hlevkin.com/06testimages.htm. [15] “Set12.” DnCNN/Testsets/Set12, Github, 13 Aug. 2016, github.com/cszn/DnCNN/tree/master/testsets/Set12. [16] Gallagher, Thomas A., Alexander J. Nemeth, and Lotfi Hacein-Bey. "An introduction to the Fourier transform: relationship to MRI." American journal of roentgenology 190.5 (2008): 1396-1405. [17] Naveen, S., and V. A. Aiswarya. "Image denoising by Fourier block processing and Wiener filtering." Procedia Computer Science 58 (2015): 683-690. [18] Dabov, Kostadin, et al. "Image denoising by sparse 3-D transform-domain collaborative filtering." IEEE Transactions on image processing 16.8 (2007): 2080-2095. [19] de Cheveigné, Alain, and Israel Nelken. "Filters: when, why, and how (not) to use them." Neuron 102.2 (2019): 280-293. [20] Mark Nixon, A. S. "Feature Extraction & Image Processing for Computer Vision." (2012). [21] Chen, Wai Kai. The electrical engineering handbook. Elsevier, 2004. [22] Canny, John. "A computational approach to edge detection." IEEE Transactions on pattern analysis and machine intelligence 6 (1986): 679-698. [23] Romeny, Bart M. Haar. Front-end vision and multi-scale image analysis: multi-scale computer vision theory and applications, written in mathematica. Vol. 27. Springer Science & Business Media, 2008. [24] Brown, Robert Grover, and Patrick YC Hwang. Introduction to random signals and applied Kalman filtering. Vol. 3. New York: Wiley, 1992. [25] Khalil, Hassan K., and Stephanie Priess. "Analysis of the use of low-pass filters with high-gain observers." IFAC-PapersOnLine 49.18 (2016): 488-492. [26] Tanner, Darren, Kara Morgan-Short, and Steven J. Luck. "How inappropriate high-pass filters can produce artifactual effects and incorrect conclusions in ERP studies of language and cognition." Psychophysiology 52.8 (2015): 997-1009. [27] Singh, Parminder, and Harjinder Singh. "A comparison of High Pass Spatial Filters using Measurements and Automation." Published in IJERT 1 (2012): 50. [28] Dogra, Ayush, and Parvinder Bhalla. "Image sharpening by gaussian and butterworth high pass filter." Biomedical and Pharmacology Journal 7.2 (2014): 707-713. [29] Butterworth, Stephen. "On the theory of filter amplifiers." Wireless Engineer 7.6 (1930): 536-541. [30] Reddy, B. Sai Tejeswar, and Valarmathi Jayaraman. "Application of Wiener Filter Making Signals Orthogonal." 2019 International Conference on Vision Towards Emerging Trends in Communication and Networking (ViTECoN). IEEE, 2019. [31] Redmon, Joseph, et al. "You only look once: Unified, real-time object detection." Proceedings of the IEEE conference on computer vision and pattern recognition. 2016.

Extracting Rib Structure from Two-dimensional Chest X-ray using High Dynamic Range Imaging Method in OpenCV Jimin Han Daewon Foreign Language H.S. in the Republic of Korea hanjimin0826@gmail.com

Sumin Han Korean International School in the Republic of Korea sumin0524@hotmail.com Abstract Background: X-ray radiography is a common clinical diagnostic technique. Single-projection chest X-ray images are frequently used to investigate the human chest structure. Interpretation of chest X-rays requires a lot of experience because, in a single projection, various structures such as lungs, heart, spine and ribs are superimposed in a single two-dimensional image. In particular, the rib bones overlap with other chest structures making it difficult to distinguish them. Purpose: This study is designed to produce a modified chest X-ray image that makes the rib bones more visible. To this end, we use various post-processing methods of the OpenCV library. Methods: We have applied post-processing to some chest X-ray images with the goal of making the rib bones more prominent. The following methods were used: histogram equalization, calBackprofect, Laplacian filter and Gaussian blur filter. Computational tools to implement these filters can be freely accessed in the OpenCV library. The quality of the enhanced images was quantified by the voting and Harris corner detection methods, and a highdynamic range image in which the rib bones are prominent was obtained by using the generative adverse network method and applying the above filtered images. Results: The lung shadow was significantly reduced in the histogram equalization and Laplacian filer postprocessed images. Also, as a result of processing, the HDR images algorithm using these images, the chest Xray image which distinguish the ribs well was finally implemented. Conclusion: The HDR imaging technique using various post-processed images can produce an enhanced imaged of ribs in a chest X-ray.

1. INTRODUCTION Image processing is concerned with working with already existing images to remove noise, segment regions of interest, etc. The recent advances in information technology and computing power have motivated increased interest in medical image processing. Among numerous types of computational processing, image enhancement plays an especially important role since it is one of the preliminary steps that are taken before starting image analysis. Image enhancement is widely used in X-ray, computed tomography (CT), and magnetic resonance imaging (MRI), where various image processing technologies have recently been applied to improve the quality and information content of images. The sharpness and quality of the image determines the accuracy of diagnosis. It is known that two-dimensional image, especially the chest x-ray image, is inaccurate in the determination of bone structures due to interference with surrounding complex lung tissue. It is known that detection of fractures or evaluating the abnormal bone structure in chest x-rays is difficult and the diagnostic rate of rib fracture using chest x-ray is about 50%.1 Therefore, additional scans such as computed tomography (CT) may be needed for a more reliable diagnosis. However, these tests put the patient at risk of exposure to more radiation. If a specially processed chest x-ray can easily distinguish the bones, one can easily evaluate the bone structures without performing additional diagnostic tests. This may reduce the risk of patient exposure to radiation to diagnose a fracture. Therefore, if post-processing can be more easily applied to detect the bone structures in a chest x-ray, it would be possible save the time and reduce costs, which are associated with using additional diagnostic tools.

The purpose of this study is to investigate whether the post processing methods such as histogram equalization, calBackproject, Laplacian filter and Gaussian blur filter are effective for extracting the bone structures from chest x-ray images. We will utilize Intel’s Open Source Computer Vision (OpenCV) library of image processing tools.2 2. MATERIALS AND METHODS This study was performed by using author’s own chest x-ray. This image is formatted as a JPG file of 153Kb in size. No other images were used in this study. In this study, several image post processing algorithms in openCV were used to transform the image, and among them, the images which the ribs were clearly visible, and the noise were reduced were selected and used for the final HDR algorithm. 2.1. Intensifying the black color by histogram equalization of the X-ray image Histogram equalization is used to improve the contrast in images by spreading out the most frequent intensity values. It accomplishes this by effectively spreading out the most frequent intensity values, i.e., by stretching out the intensity range of the image. This method usually increases the global contrast of images when its usable data is represented by close contrast values.3 This allows for areas of lower local contrast to gain a higher contrast. As a way of assessing the image transformed by this method, we used voting method for two images using previously reported method. Briefly, five independent evaluators were asked to vote on where the rib images were more clearly visible, and an assessment of the image clarity was conducted. The independent evaluators did not know what post processing was performed for both images.4 2.2. Extracting rib images by using calcBackproject Calcbackproject in OpenCV is an algorithm used to divide images or to find areas of interest in images.5 This is mainly used in the background projection method of histogram. Briefly, this algorithm can find a pixel that is probabilistically similar to the pixel we want and as a result we can extract specific images. The rib images can be selected and magnified for further examination of the equalized image by using this method. If we can segment out the rib image from the chest x-ray, we postulated that a more detailed evaluation of rib can be performed. We verified the extraction of ribs from the chest x-ray using calcbackproject algorithm. 2.3. Laplacian filter and the Gaussian Blur comparison The Laplacian filter computes the second derivatives of an image. By doing so, it tests whether a change in adjacent pixel values is due to an edge or continuous progression.5 Since Laplacian detects an edge by summing the differences between a center pixel and its neighboring pixels in four or eight directions, it is very sensitive to noise and tends to detect false edges. Therefore, it is desirable to remove the noise from an image before applying the Laplacian filter. The Gaussian filter is one of the most widely used blurring filters.7 When applying the Gaussian filter, the current pixel value is replaced by the weighted average of the neighboring pixel values. The closer to the current pixel, the larger is the weight. The premise behind the Gaussian filter is that the intensity of the pixel changes slowly so that the neighboring pixels have similar values. In contrast, noise is typically uncorrelated and is therefore suppressed in the weighted average. 2.4. Harris corner detector to compare the two filtered images To compare the Laplacian filter process and Gaussian blur filter technique, we used Harris corner detector for both images. Harris corner detector is a corner detection operator that is commonly used in computer vision algorithms to extract “corners” and infer features of an image.8,9 If noise is reduced in the chest x-ray image and the distinction between each part of images in chest x-ray becomes apparent, this method may easily find corner points where rapid image changes occur. This is one of the ways to find a specific part of an image that has become clearer due to reduced noise. Using this method, we looked up which of the Laplacian filter or Gaussian blur filter find the corner points well. 3. RESULTS 3.1. Histogram Equalization After performing the histogram equalization, the contrast enhanced image was obtained (Figure 1). As a result of the evaluation of the images, five of five (100%) people agreed that the distinction between rib images in post-processing images was easier and confirmed that this method effectively distinguishes ribs. The image accomplishes this by effectively spreading out the most frequent intensity values, i.e. stretching out the intensity range of the image. This method usually increases the global contrast of images when its usable data is represented by close contrast values. This allows for areas of lower local contrast to gain a higher contrast.

Fig. 2 displays the image histograms before and after equalization. It can be seen that the distribution of intensity values in the image pixels has been made more uniform after the equalization. In summary, the histogram equalization method has improved the global contrast of the image and allowed for areas of lower local contrast to become more visible.

Figure 1. The chest X-ray image before (left) and after (right) histogram equalization. In the right image, the contrast between white bones and the black background is increased. As a result, lung shadow was reduced, resulting in prominent rib shadow.

Figure 2. The histogram of the image shown in Fig. 1 before (hit1) and after (hit2) equalization. The horizontal axis of the histogram is the intensity value, and the vertical axis represents the frequency at which it appears in the image. The original histogram had small values for pixel intensities between 0 and 100. The equalized histogram is much more evenly spread. 3.2. Extracting ROI by using calcBackproject As one can see in the Fig. 3, the red line (H1) shows the color distribution of the entire X-ray. The blue line (H2) compared the color distribution of region of interest (ROI) (ROI = (148, 156, 124, 201). By using this method, it is possible to identify abnormalities in a particular region or color distribution in a particular region.

Figure 3. Color distribution curve of a specific part of the x-ray image 45

This is the color distribution curve when a specific part of the X-ray image is extracted with calcBackproject in Open CV. The X-axis represents the brightness intensity and the Y-axis represents the number of pixels. H1 is the color distribution curve of the entire area and H2 is the color distribution curve of the specific area, in other words the region of interest. To view images, go back to Figure 4, the H1 graph shows highly condensed color intensity values in the 100-150 ranges, where as in the H2 graph, the color distribution is more widely spread, ranging from 50-150. This means that there is higher contrast within the image that is extracted, and the color range is more tone-down.

Figure 4. Equalized image (Left) and calcBack Project image (Right) after extracting the ROI (ROI =(774, 296, 600, 536)) 3.3. Laplacian filter and the Gaussian Blur Comparison The Laplacian filter and the Gaussian Blur are both applied in order to compare which filter is most effective in mitigating noise without degrading the original image quality. It can be clearly seen that the Gaussian blur somewhat degraded the sharpness of the image, while the Laplacian managed to maintain the resolution as well as reduce noise. (Figure 5) However, the difference between the two post process images is hard to distinguish. For this reason, the harries corner points detection method was used to accurately distinguish the quality of two images.

Figure 5. The images before (left) and after processing of the Laplacian filter (center) and Gaussian Blur filter (right). 3.4. Harris corner detector to compare the two filters By using the Harris corner detection method, we investigated which filter allows for better corner detection. ‘R’ is an alphabet which means the right side in the chest x-ray and is an independent image artificially inserted into the image to indicate the direction. It is the most clearly distinguished image in the chest x-ray image, and we have identified it as a target to see if the algorithm distinguishes ‘R’. As a result, the ‘R’ on the top left corner of Gaussian blurred image could not be detected, whereas the ‘R’ of the Laplacian filtered image could be detected. This shows that the Laplacian filter produces a better image quality (Fig. 6) than Gaussian blur filtered image. Briefly, Harris Corner detection has preserved "R" in the image converted to Laplacian filter, but the converted image to Gaussian blur filter resulted in an 46

error in which "R" was expressed as noise. This result shows that the method of changing Laplacian filter has higher clarity.

Figure 6. The images after processing the Gaussian blur filter (left) and Laplacian filter (Right). After performing the Harris corner detector algorithm, the “R” not found in the left image is recognized as corner points (red cross) in the right image. 4. DISCUSSION To this day, single projection X-ray images remain an important diagnostic tool. A well-known limitation is that single projections are unable to resolve overlapping structures, making it difficult to assess or visualize certain threedimensional features. However, it can be expected that this limitation can be overcome by using appropriate image processing filters that reduce noise and enhance the visibility of various features. In this paper, we have discussed a variety of methods including histogram equalization, Laplacian filter, and Gaussian blur. In the case of histogram equalization, the range for color intensity had been widened, as a result creating higher contrast. Histogram equalization is an image enhancement method that calculates pixel values of an output image by using the cumulative distribution as a transformation function. Histogram equalization transforms a low-contrast input image with a narrow range of pixel values into a high-contrast output image with a wider range of pixel values. In other words, if the distribution of pixel values in the original image is narrow and concentrated around a certain value, either dark or bright, histogram equalization produces a clearer image with a wider distribution of pixel values. The Laplacian filter was effective in enhancing the sharp discontinuities.10 The Gaussian blur resulted in considerable noise reduction but the overall image quality had been degraded and the edges became more opaque. This difference was proved by Harris corner detection method. A corner is a point whose local neighborhood stands in two dominant and different edge directions. Edge points are the point of rapid change in one direction and the points without change are called flat points. This method was used under the assumption that the distinction between images could be difficult if the noise was severe. As a result, it was confirmed that the difference between the two images was meaningfully distinguished. Laplacian filter showed better noise reduction after postprocessing the chest x-ray. The region of interest (the areas of potential abnormalities to be diagnosed) can be selected and magnified for further examination of the equalized image. Because it is difficult to distinguish a specific area from the whole X-ray image, a more detailed diagnosis can be performed by segmenting out only a specific range. As a result, the resolution quality enhances significantly. In addition, the reader also takes an important role in displaying the modified images. The X-ray can be deciphered only when the reader has the corresponding capacity to express such high-resolution images. While conducting this research, when analyzed on a high-resolution computer, the image quality was not impaired, but on a computer screen with a lower resolution, the resolution was conspicuously degraded. In other words, although brightness control and filtering methods are important factors in deciphering images of X-rays, before that, a high-definition reader that is capable of reading high-definition X-rays are equally, or even more important. In summary, we were able to obtain some satisfactory enhancement of bone images in a chest x-ray by using various methods for image post-processing. However, we only succeeded in extracting bone structures in one specific 47

image. The future studies should address a wider and more statistically significant set of images including bones and soft tissues. Although open source programs (OpenCV) that are publicly available were used in this study, it is expected that the research on advanced image processing techniques will be developed in the future by developing artificial intelligence or improved image acquisition processes.

REFERENCES 1.

https://emedicine.medscape.com/article/825981-overview

http://www.OpenCV.org

https://docs.opencv.org/3.4/d4/d1b/tutorial_histogram_equalization.html

SA van Zyl et al., Comparison of the perceived image quality between two digital imaging systems for neonatal bedside radiography - A case study, Radiography 21(1), 2015, e45-e49

https://docs.opencv.org/3.4/da/d7f/tutorial_back_projection.html

https://www.harrisgeospatial.com/docs/LaplacianFilters.html

https://docs.opencv.org/master/d4/d13/tutorial_py_filtering.html

https://medium.com/data-breach/introduction-to-harris-corner-detector-32a88850b3f6

https://docs.opencv.org/master/dc/d0d/tutorial_py_features_harris.html

10. https://www.sciencedirect.com/topics/engineering/laplacian-filter 11. https://bskyvision.com/703

Inverse Problems and Noise Reduction in Imaging by Post-Image Processing Sumin Han Korean International School in the Republic of Korea sumin0524@hotmail.com

I. INTRODUCTION Appearance of noise in biomedical images can have severe detrimental effects. Especially when probing CT, MRI, and X-Ray imaging, a simple mistake while decoding can bring uncertain judgements, later giving significant backlashes. However, by processing the images of the patient with various filters and methods, it is possible to meet the goal of performing the most appropriate tomography. Then, what is this image processing? Image processing is the study of researching and analyzing images. In images, there are various hidden information that are processed unconsciously. The key to image processing is to extract useful information of image features, such as edges, objects, and contours, by using different attributes of the image, including color, pixel-to-pixel relationships, object placement, and other fine details.

Figure 1. Fourier transforms frequency Repetitive scans can be harmful for people who take x-rays, CTs, and MRIs due to excessive exposure to radiation, creating a problem with re-shooting. To solve these problems, there are many ordinary methods of fitting the position and height of radio waves in Fourier transform, which is a mathematical formula that associates a signal sampled in time or space with the same signal sampled at frequency. This method can be usually done by using “frequency modulation”, “phase modulation”, and “amplitude modulation.”1 Furthermore, in signal processing, fourier transforms play a role in representing important characteristics of the signal-frequency components. By exclusively recognizing and using the method of Fourier transform for two dimensional images, we are able to point out the noise in raw images which later enables controlling the wavelength and, ultimately, reaching the conclusion of reducing noise. However, the weakness of this method can be seen that even if a normal picture is printed, it is difficult to detect a slight error in two-dimensional space. In other words, we can tell what happens on a signal, not when it happens. Thus, the outcome required the most convenient and effective method to perform a maximum quality. In doing so, by using the 1

Google Image Noise Reduction by Frequency 79,000,000 Results

common computer Bichon, especially PILLOW; an open source library; and OpenCV (Python and C++) with ANN (Artificial Neural Network), I came to think of a post-processing way to reduce noise and to consider how to proceed with it. II.METHOD AND DATA OUTPUT CT images showed stroke, but because it was difficult to determine if there was a problem with the brain, I took an MRI. By comparing and analyzing these two photos, we want to check how much noise reducing in CT and MRI images is possible and to see if these methods are effective. Currently, the typical way to research post-image processing of computer vision is PILLOW and openCV using Python and C++. I'm going to use these two methods together. The reason is that Pillow is simple to use in a more intuitive pre-processing and openCV in the Artificial Neural Network (ANN). The PILLOW library allows simple pre-processing, which can be adjusted for image expansion, brightness, and color arrangement. With this, we are going to print out the first processing image. The independent variable at this time is, of course, the brightness adjustment and the dependent variable the calamage dimension printed for the brightness adjustment. 1) First Step - Enhanced Image Image processing of CT tomography is not just a method of reducing noise but also a method that is used a lot in post processing to identify problems in fine parts and to help diagnose them. The brightness enhancement filter was utilized in order to demonstrate the minute details of the raw CT tomography and, ultimately, to reinforce the overall quality of the image. First, using the python code, the raw CT image was brought and opened to further boost the quality. Then, the process of adjusting the brightness in order to augment the quality of the tomography was implemented. Especially here, after bringing the raw CT tomography, 40% of the brightness was increased by the enhancer.

Figure 2. Raw image

Enhanced image Source: origin and mapping by python with Pillow

The two images above show the distinction between the first original tomograph and the enhanced tomography after applying the previous code. 2) Second step - Enhanced Image Noise Reducing & Image Crop After producing an enhanced image, median filtering and Kernel were used to reduce the noise. In other words, the median filter—a type of nonlinear filter—and the Kernel filter allow the communication of the color distribution and set of color values. These filters are especially appropriate for decreasing “salt and pepper” noise, which in other words, replaces the gray level of the pixel by the estimation of median of the gray levels instead of the average operation. For this process to be executed, we need to specify the kernel size, list the pixel values covered by the kernel, and average the two median values.

Here, the set Kernel filter’s color values were (3,3),[1,1,-1,3,1,2,1,1,1]. This set of color values adjusts the color density by handling the noise produced particularly in the range of the gray scale. At the same time as managing the kernel filter, the surrounding noises tend to match the enhanced image’s median value, which in this method is three. Additionally, the enhanced image below is cropped into its ROI (range of interest), which in this case is the area of the tumor.

Figure 3. Noise reducing

roi = (292, 234, 212, 284) ROI (range of interest) ex) tumor

3) Third Step - KAZE & AKAZE P. F. Alcantarilla et al. put forward KAZE features in 2012 that exploit non-linear scale space through non-linear diffusion filtering. This makes blurring in images locally adaptive to feature-points, thus reducing noise and simultaneously retaining the boundaries of regions in subject images. Then, P. F. Alcantarilla et al. presented Accelerated-KAZE (AKAZE) algorithm in 2013 2[4], which is also based on nonlinear diffusion filtering like KAZE— but its non-linear scale spaces are constructed using a computationally efficient framework, Fast Explicit Diffusion (FED). More explicitly, the AKAZE algorithm has more distinctiveness at diverse scales due to their nonlinear scale spaces. In other words, in the nonlinear scale constructed by KAZE, the feature points are detected by nonlinear diffusion filtering, and the descriptor is calculated. Nonlinear diffusion filtering especially enables noise reduction smoothly ‘within’ regions defined by object boundaries as well as image contour enhancement. By using a local adaptive blurring method, noise is removed, and details such as object boundaries are maintained to improve the accuracy of feature point detection and performance. As a result, in the perspective of the recipient, the outcome obscures the desired information and rather depicts undesirable products. In other words, the image appears ambiguously, and only the text signal seems to be revealed. Initially, the code used for this particular CT tomography introduces the Gaussian blur filtering, which adjusts the surrounding color value to the average of the overall color distribution of the CT image. Here, the Gaussian Blur is calculated as (cv2.GaussianBlur(gray, (5, 5), 0.0). Then, with cv2.KAZE_create()and cv2.AKAZE_create(), kazeF and akazeF are produced. When each kazeF.detect(gray)and akazeF.detect(gray)detects the key points in kp, the feature point filtered_kp is displayed as a red circle in dst1. Using the feature's degree and angle, the rotating rectangle (in green) and circle (in blue) were drawn in dst2. More explicitly, the code of the colors are described as color=(R,G,B), where red is color=(255,0,0)in KAZE; green is color=(0,255,0)for polylines in AKAZE ; blue is color=(0,0,255)for circles in AKAZE.

P. F. Alcantarilla et al., "Fast explicit diffusion for accelerated features in nonlinear scale spaces", British Machine Vision Conference, 2013.

Figure 4. KAZE

AKAZE Source: photo by origin with python

4) Fourth step - MRI Image Processing A similar method with MRI imaging was used to process the tomograph of MRI as well after bringing the raw image by utilizing the python code Image.open("MRI.jpg"). Then, the quality of the tomography was boosted through the process of adjusting the brightness. Particularly for MRI, after bringing the raw MRI tomography, 20% of the brightness was increased by the open source code. The most appropriate adjustment in the open source code was enhancer.enhance(1.4).show()for the maximum quality because enhancing the image by 40% brightness brought an outstanding result.

Figure 5. Origin Brain MRI

Enhanced image processing Source; Origin data Enhanced with python

5) Fifth Step - Enhanced Image Noise Reducing & Image Partial The noise appearing in the image can be seen as blurriness. When reducing the noise of the brightness-adjusted image (brightness and contrast filter), we found out whether the image in the MRI made it easier to understand where the lesion occurred in the first place.

roi= (595, 2, 0, 0) Figure 6. Enhanced Image Noise Reducing

Partial image

III. DATA RESULT AND PHOTO RESULT 1. Figure 1. Plot data and change to two-dimensional Even if the brightness is adjusted in two dimensions, it can be seen that there is no difference in the two dimensions of the histogram.

Figure 7. Origin photo grayscale hist

Enhanced photo grayscale hist

2. Noise Filter Grayscale Hist Partial Image Hist - Enhance As noise is filtered, changes in the histogram other than the enlarged histogram area are too low in color compared to the figures, making it difficult to recognize visual changes.

The singularity output is determined to reveal various problems when the doctor looks at a specific part of the image. It would be better to find a better way than this, but it would be possible to line up by marking off areas. KAZE and AKAZE Data Output in slen(kp)= 1338, len(filtered_kp)= 604, des.shape= (604, 64), des.dtype= float32, des= [[ 1.8235341e-03 -3.4456082e02 6.9455267e-03 ... -1.6365859e-05, 2.8637758e-05 1.6365859e-05], [-2.4228254e-02 1.0984519e-02 4.0812206e02 ... -2.6736813e-02, 1.1214775e-02 2.6736813e-02], [-7.2120025e-07 -9.6736130e-07, 7.2578786e-07 ... 2.6060322e-02, 8.3298288e-02 9.5527194e-02], ..., [-9.1795690e-02 -9.7255811e-02 , 9.3765698e-02 ... -3.5151705e02, 5.3944852e-02 7.7238247e-02], [ 1.8907750e-01 3.4133333e-01 3.1967568e-01 ... 2.2708733e-02, 3.9469622e02 6.7014717e-02],[-3.0515403e-03 3.0499808e-02 2.9456826e-02 ... 3.9162584e-02,3.3856373e-02 5.0596599e02]] MRI data histogram Origin MRI hist

Noise reducing histogram & partial Noise Reducing Histogram (entire)

Enhanced hist

Partial Histogram

IV. DISCUSSING DATA OUTPUT AND RESULTS Comparing the above CT and MRI pictures, CT has the same two-dimensional color space, so it is more difficult to interpret lesions than from an MRI, which takes diagnosis of lesions closer to two-dimensional space due to adjusting color brightness or simple noise rehydration. I think it is also worth considering as an alternative to proceed with advanced adjustment in the future. The reason is that, assuming CT has a high probability of noise due to the high color of Gray Scale, I would like to suggest a review of the fact that this color can be changed to two dimensions of Black and White to reduce noise or use KNN technology. And MRI shows that disease area discrimination can be easily carried out with simple postprocessing in-hand Fourier Transform. The results of the experiment also show that the lesions can be accurately identified if the secondary noise rehydration and photographic segmentation are performed, and the analysis of the specific points of histogram KAZE and AKAZE show that the lesions are unique when the CT scans are expanded. 54

Furthermore, it is useful to review the expansion of various regions. It is judged that the help in diagnosis is not a specialist who has undergone a high degree of practical training; it can expose analytical problems in the eyes of the general public. Although the results of this experiment show that simple noise processing techniques can be clearly distinguished by simple 1.3 - 1.4 fold control and noise reduction alone if applied to MRI in two-dimensional space, it is judged that the MSE method can be used to deduce appropriate brightness. V. CONCLUSION – ACHIVEMENT AND LIMITATION Although this study of noise processing is unfortunately limited by the short preparation time and the use of other people's medical information at will, due to the nature of the Korean Medical Law, the method of noise processing is only fragmentary image processing of the same person, CT, MRI of the same person, so we can suggest a review opinion on noise reducing of CT and MRI. 1. In the case of CT images, it was determined after MRI data analysis, and it was judged that the noise rehydration method is more efficient by adjusting the enhanced adjustment to the dark side and bringing the gray color closer to white and black. 2. In the case of CT using the grayscale data adjustment method of MNIST3 data, which is commonly used in machine running4 and deep running, we could see that grey can be more helpful for noise reduction if we judge noise and finely adjust it to black and white. 3. In the case of MRI, better results were obtained using the method of noise re-during 1.3 to 1.4 times without the need for complex post-processing processes. 2. In the future, more simultaneous CT and MRI processing of the same person could help us find improvements in the post-processing of lesion diagnosis for CT scans. It is most regrettable that there is a lack of data to derive the results. In particular, if there are various cases of CT photographs, it can be judged that the method to produce the results of the analysis could be inferred more specifically.

REFERENCE 1. Studying to openCV programing With python 2. OenpCV4 using C# and Python 3. Studying to Computer Vision with Python 4. A contribution to the problem of correct description of electromagnetic waves in layered media with magnetic properties 5. P. F. Alcantarilla et al., "KAZE features", European Conference on Computer Vision, pp. 214-227, 2012. 6. P. F. Alcantarilla et al., "Fast explicit diffusion for accelerated features in nonlinear scale spaces", British Machine Vision Conference, 2013. 7. https://docs.opencv.org/master/d3/d61/classcv_1_1KAZE.html 8. https://docs.opencv.org/3.4/d3/d61/classcv_1_1KAZE.html 9. Akaze:Pablo F Alcantarilla, Jesús Nuevo, and Adrien Bartoli. Fast explicit diffusion for accelerated features in nonlinear scale spaces. 10. Trans. Pattern Anal. Machine Intell, 34(7):1281–1298, 2011. 11. http://yann.lecun.com/exdb/mnist/ THE MNIST DATABASE of handwritten digits 12. https://scikit-learn.org/stable/ 13. https://pytorch.org/ 14. Expert-level detection of acute intracranial hemorrhage on head computed tomography using deep learning Weicheng Kuoa , Christian Hӓnea , Pratik Mukherjeeb , Jitendra Malika,1, and Esther L. Yuhb,1 a Electrical Engineering and Computer Sciences, University of California, Berkeley, CA 94720; and b Department of Radiology and Biomedical Imaging, University of California, San Francisco, CA 94107 Contributed by Jitendra Malik, September 11, 2019 (sent for review May 10, 2019; reviewed by Jack W. Tsao and Andrew Zisserman

3 4

http://yann.lecun.com/exdb/mnist/ THE MNIST DATABASE of handwritten digits https://scikit-learn.org/stable/ 55

A Study on the composition of certain antiseptic and antibacterial bacteria and Ginseng Fermented Ginseng Solution Ji Soo Hwang Williston Northampton School in the U.S. jisoo120701@gmail.com ABSTRACT Fermented ginseng and 6-year-old red ginseng fermentation solutions have been demonstrated in many experiments to enhance human immunity, such as saponin and ginsenoid, and to have a good effect on the human body [. It has been interested in a composition that is not harmful to the human body and can be used by anyone by preserving the ingredients in the ginseng fermented liquid and adding antibacterial and anti-inflammatory substances. However, in the case of these structures, the study showed that substances are often combined or deformed, and I wanted to study the substances that did not undergo such deformation. In addition, it was thought that it would be best suited to the present era if the composition added a function that fosters some of the antibacterial capabilities in ginseng and the anti-inflammatory function. In addition, the study was conducted in a way that does not have a unique smell from fermentation in consideration of people who are sensitive to smell, and experiment was conducted to find substances that had the ability to remove E. coli and staphylococci separately from ginseng effects. This experimental method used the disk experiment method and the ginseng extraction separation method by centrifuge, and after removing bacteria from the fermented ginseng extract by ultraviolet radiation, it was confirmed that the bacteria and ginseng extracts function as separate substances. However, many of the main experiments and studies conducted detailed experiments to extract certain substances were regrettable that the combination of bacteria and bacteria after the extraction of certain substances became a substance that high school students could not handle. KEYWORDS: Enlargement of the immune system, antiviral, ginseng extract, Actinobateria, effect, efficacy

1. INTRODUCTION Recently, anti-virus treatments have been spotlighted due to COVID-19, while antibacterial substances such as antibacterial tissue, anti-bacterial film, and anti-bacterial soap that help prevent viral infections are also receiving much attention. [1] In particular, lung-related preventive medicine is increasing in importance and deaths from lung diseases are increasing.[2] When anti-inflammatory and antibacterial composition materials are created by mixing ingredients that have proven antimicrobial effects independently, the main purpose was to effectively remove viruses, which are the main factors of the disease, and to find substances that increase the effectiveness of existing substances in ginseng extract. And bacillus exists in many Korean soil and has been proven to be effective as a water source to curb the growth of viruses.[3] However, it was not the purpose of the study to cultivate simple bisexual bacteria, but it was believed that substances with anti-inflammatory lung disease and anti-cancer properties could prevent transmission. The study on antibacterial substances began with research on ginseng, inspired by ginseng concentrates, which Koreans consume a lot as supplements. The ingredients of saponin (Saponin), ginsenoside (20(R) and 20(S)ginsenoside-Rg3) contained in fermented ginseng and red ginseng fermentation for 6 years have been proven to have anti-cancer and antioxidant effects as well as strengthening human immunity.[2]A study on the compositional substances that can enhance immunity and achieve various effects by adding antibacterial and anti-inflammatory substances to the ingredients in the ginseng fermented liquid was started. In this study, we conducted an antimicrobial experiment to visually confirm that the inoculated fungus does not spread and disappears to ensure that the antibacterial effects of ginseng widely known as immune-boosting and actinobacteria, which is known to have antimicrobial effects. [4] 2. MAIN BODY 2.1. Side effects and causes of antibiotics

The use of natural antibiotics available to anyone to eliminate side effects from chemicals is becoming a new challenge for the medical community. Recently, it has been recognized as a health-functioning food. 2.2 Cause of pulmonary tuberculosis and raw materials for early antibiotics Actinobacteria lives in our bodies as well as in nature such as soil, rivers, and lakes. [4] Actinobacteria is already widely used in the medical community as a derivative of secondary metabolism. Streptomycin, Terramycin, and Aureomycin antibiotics with various effects can be extracted. The reason why many antibiotics are derived from Streptomycycles is because of the large DNA composition of Streptomycoses.[5] 2.3 Effects of Ginseng Peptide is used to restore and activate damaged cell tissue while at the same time it is used to root out the disease. Peptide is a substance that mediates a form of amino acids and proteins. The molecular weight is smaller than protein, the molecular weight is larger than amino acids, and the nutritional content of proteins is incomparably abundant. It is the most important active substance that plays an important role in the process of growth, metabolism, disease, aging and death in the human body. 3. METHODOLOGY 3.1 Experiments and results extraction Streptomyces griseus subsp. griseus (KCTC 1080) In Seed Culture Media, add 4g of yest extract per liter, 10g of maltose extract and 4g of glucose, and incubate it under the conditions of pH 7.2.

Figure1. Fungal incubate

Seed Culture Media: YM broth

Figure 2. Before incubation, test group (right test group: additional ginseng) and control group (left control group)

As can be seen in Figure.3, the test group (right) in this experiment is S.griseus + Ginseng (40g) and the control group (left) is S.griseus + media. The two Bennett's media were incubated for 54 hours, the same as the previous Streptomyces incubation conditions. 3.2 Results after incubation

Figure 3. Classification after incubation After incubation for sample preparation, only suernatant of the results of a 30-minute centrifuge using a medical thermostatic centrifuge (LaboGene 1580R) was taken and used as an antimicrobial experimental sample.Prior to the fermentation of ginseng, homogenized ginseng was sterilized with a grinder to exclude the effects of other bacteria other than the fermented bacteria (S.griseus).

Figure 4. Results of disk-based antibacterial experiments As a result of the antimicrobial experiment, as seen in Figure 5, the ginseng fermentation group (+) marked as 'Con' in the A.soli target strain plate and the ginseng-free group marked as (-) with (+) antimicrobial antigen. CONCLUSION As a result of this experiment, it was assumed that a variety of products could be extracted using this extract by using the same extract as the ginseng extract, which proved to be immune to the human body, and the initial extraction of the waste solution, and the same extraction of Streptomyces griseus subp.greiseus-KCTC 1080. With foreign substances, we succeeded in extracting possible liquids with natural scents that are good for the body and good for the scent, air fresheners used in the house, or product ingredients that remove odors, remove germs, and help the body. 58

FUTURE IMPROVEMENT TASKS In addition to known substances such as saponin and ginsenoid, Streptomyces griseus subp.griseus-KCTC 1080 is a more granular data in the extraction process. It is also a challenge to study the effectiveness and effectiveness of the material if each function of is maintained.

REFERENCES [1] A Study on the Manufacture and Characteristics of Antimicrobial Film Using Water-soluble Polymers”, Korean Chemical Engineering Research, 49, 4, pp.449-455, 2011 [2] Statistics for the causes of death in 2019 show Statistics Korea at September 21.15:00 [3] Mochizuki, M. et al.“Inhibitory Effect of Tumor Metastasis in Mice by Saponins, Ginsenoside-Rb2, 20 (R)-and 20 (S)-Ginsenoside-Rg3, of Red ginseng”, Biological and Pharmaceutical Bulletin, Vol. 18, Issue9, pp.1197-1202, 1995 [3] Anandan, R. et al. “An Introduction to Actinobacteria”, DOI: 10.5772/62329 [4] Procopio R.E.L. et al. “Antibiotics produced by Streptomyces”, The Brazilian Journal of Infectious Diseases, Vol. 16, Issue 5, pp.466-471, 2012.10 [5]NCBI, Streptomyces Grisesus, https://www.ncbi.nlm.nih.gov/Taxonomy/Browser/wwwtax.cgi?mode=Info&id=1911&lvl=3&keep=1&srchmode =1&unlock&mod=1&log_op=modifier_toggle [6] U.S. Center for the Practical Use of Pharmacopoeia Cell Therapy and Genetic Therapy www.celltherapy.re.kr [7] https://www.slideshare.net/EditageKorea/ss-32115293