IJSRD - International Journal for Scientific Research & Development| Vol.4, Issue 04, 2016 | ISSN (online): 2321-0613
Emotion and Color Combination Based Aesthetic Classification of Images Mitali Paygude1 Ankita Tapadia2 Purvi Lahoti3 Prof. M.V.Phatak4 1,2,3,4 Department of Computer Engineering 1,2,3,4 M.I.T Pune Abstract— Aesthetics in the world of art and photography refers to the study of beauty and taste. However, determining aesthetic qualities of any visual representation of art is a highly subjective task. Hence, there is no unanimously agreed standard for measuring aesthetic value. In spite of the lack of formed rules, certain features in photographic images are believed, by many, to please humans more than others. Image processing involves taking image as an input and performing operations on it. The main aim of our system is to tell if certain image is generically appealing or not based on the color aesthetics of the image. In this system we take an image from a training data set of images, extract color features of that image and based on the color combination and contrast ratio between the foreground and background of the image, classify the image as appealing or not. We also took a few surveys for understanding the color combinations which are more preferred by humans and also the variation in the color preferences according to gender along with emotions which were evoked. The survey results were then used as guideline in designing the classifier rules. This paper will depict the work done on this system in brief. Key words: Aesthetics, Classification, Object detection, Color contrast I. INTRODUCTION Most available systems for image classification and Content Based Image Retrieval are focused on object and scene recognition. The core idea of the system is to identify and classify a given image as aesthetically appealing or not, based on the color combination between the foreground and the background of the image. The system also tags every image with an emotion that is generally evoked in human beings after looking at such images. This system also classifies images according to gender preferences. Some work has been done in this field pertaining to this topic. It is mentioned in the next sub section. II. PREVIOUS WORK The main aim of our system is classification of images on the basis of color contrast. Two similar systems were available, however with few drawbacks. XEROX Quality Image Search performs automatic image categorization to first understand the content of the image, then to filters the images with good and bad aesthetic elements. It however had a fixed database of images. It did not allow the user to input the images of user’s choice.
Fig. 1: XEROX ACQUINE - Aesthetic Quality Inference Engine, a publicly accessible system where users upload their photographs and have them rated automatically for aesthetic quality. The system uses a support vector machine based classifier which extracts visual features on the fly and performs real-time classification and prediction. The developers of the system have not disclosed the algorithms or features considered while classification. This system has been taken offline now. Paper Author Description Studying Ritendra Aesthetics in Automatically Datta, Dhiraj Photographic classifying images Joshi, Jia Li, Images Using a on the basis of their and James Z. Computational aesthetic quality Wang Approach Using simple statistical measurements, lie Emotion Related global image Structures in Martin Solli, histograms to Large Image Reiner Lenz classify images in Databases categories wrt emotions and aesthetics Aesthetic Quality Survey of al existing Anu Joy, Classification of approaches and Sreekumar photographs : A techniques for K Literature Survey image classification Aesthetic response to color Karen B. Understanding color combinations: Schloss & preferences , preference, Stephen E. similarity and harmony, and Palmer. combinations similarity. Table 1: III. PROPOSED WORK First we take an image as input which we have to test against the classifier. Then we detect the object using techniques like blurring and edge detection. Once the object and the background are separated, color is extracted. HSV
All rights reserved by www.ijsrd.com
696
Emotion and Color Combination Based Aesthetic Classification of Images (IJSRD/Vol. 4/Issue 04/2016/171)
values are calculated. Using the color wheel, contrast between the foreground and background is calculated. Then we attached emotions on the basis of surveys taken.
2) HSV Color Model
Fig. 3: HSV Color Model Hue represents the color type. It can be described in terms of an angle on the above circle. Although a circle contains 360 degrees of rotation, the hue value is normalized to a range from 0 to 255, with 0 being red. Saturation represents the vibrancy of the color. Its value ranges from 0 to 255. The lower the saturation value, the more gray is present in the color, causing it to appear faded. Value represents the brightness of the color. It ranges from 0 to 255, with 0 being completely dark and 255 being fully bright. White has an HSV value of 0-255, 0-255, 255. Black has an HSV value of 0-255, 0-255, 0. The dominant description for black and white is the term, value. The hue and saturation level do not make a difference when value is at max or min intensity level. 3) Color Contrast’s Effect on Aesthetics The aesthetic experience of any given color is strongly influenced by its participation in combinations of two or more colors. More the contrast, more the appeal on a given image. Colors can be divided into categories like warm or cool colors, or analogous or complementary colors. There can be different tints or shades of one color. Such different shades of colors and their contrasting colors are included in our system.
Fig. 2: Figure represent the system architecture diagram of the system A. Color Theory 1) Color Models Color vision can be processed using RGB color space or HSV color space. RGB color space describes colors in terms of the amount of red, green, and blue present. HSV color space describes colors in terms of the Hue, Saturation, and Value. In situations where color description plays an integral role, the HSV color model is often preferred over the RGB model. The HSV model describes colors similarly to how the human eye tends to perceive color. RGB defines color in terms of a combination of primary colors, whereas, HSV describes color using more familiar comparisons such as color, vibrancy and brightness. We have used the HSV color model to process our color vision.
B. Algorithms Following are the algorithms that have been used in the system. These are the preprocessing techniques for our system: 1) Blurring of an Image In image terms blurring means that each pixel in the source image gets spread over and mixed into surrounding pixels. Another way to look at this is that each pixel in the destination image is made up out of a mixture of surrounding pixels from the source image. Blurring an image reduces the sharpening effect, this makes the object detection more accurate.
All rights reserved by www.ijsrd.com
697
Emotion and Color Combination Based Aesthetic Classification of Images (IJSRD/Vol. 4/Issue 04/2016/171)
Fig. 4: Original and blurred image a) Algorithm 1) Traverse through entire input image array. 2) Read individual pixel color value (24-bit). 3) Split the color value into individual R, G and B 8bit values. 4) Calculate the RGB average of surrounding pixels and assign this average value to it. 5) Repeat the above step for each pixel. 6) Store the new value at same location in output image 2) Grayscale Conversion This is needed for the next algorithm. Sobel edge detection is performed on grayscale images.
Fig. 5: Original and grayscale image 3) Edge Detection A spatial edge detection filter that detects edges by finding the gradient of an image. This means finding out a dramatic difference between the value of pixels. Edge detection is required for number of applications such as: Shape detection. Face recognition. We have two template matrix for applying sobel algorithm -1 0 1 -2 0 2 -1 0 1 1 2 1 0 0 0 -1 -2 -1 Table 2: GxGy
Fig. 6: Original And Edges Detected Image 1) The input image is first converted to gray scaled image. 2) Traverse through entire image. 3) For each pixel in the image we will take a window of 3*3 pixel and multiply it the given template for matrix. 4) Then we will calculate the G using formula
Once these pre-processing steps are performed on the input image, we extract the RGB values of the object and the background. We convert the RGB values to HSV. We calculate the mean HSV of all the pixels in the foreground and the background. We then compare the HSV values by subtracting the values. More the value, more the contrast. Thus, the images are classified as appealing or not appealing on the basis of contrast detected in the object and the background.
Fig. 7, 8: Image with low Contrast & Image with medium Contrast C. Gender Specific Classification Now that the contrast is calculated, each image is tagged with an emotion. These emotions generally vary depending on the gender of a person. We conducted a survey to find out which emotion was invoked in which gender prominently. On the basis of that survey and few other surveys which we found online, we tagged the input images with emotions evoked. IV. CONCLUSION We have designed and implemented a system which classifies a given image or set of images based on the color aesthetics of the image/s. It tags the image with an emotion based on the same. We have put a lot of work in understanding color theory, different color models, color wheel and color psychology. We have taken surveys of which color combinations are preferred by people and what kind of emotion they associate with it. Also, how color preference vary according to gender. Our work has led us to believe much more potential in this stream of work can be possible which can change the image classifying systems used. V. FUTURE WORK The system can be extended for applications in image classification. The proposed system is a prototype of automatic aesthetic classification system and can be further extended to build systems with advanced feature extraction incorporated in it. The system can be used for image classification when the digital photographs are clicked itself. For example, incorporating this software in digital cameras. The system can be extended for multiple object images. The system can be extended to factor in userspecific attributes like age, occupation, nationality to give more accurate results.
All rights reserved by www.ijsrd.com
698
Emotion and Color Combination Based Aesthetic Classification of Images (IJSRD/Vol. 4/Issue 04/2016/171)
ACKNOWLEDGMENT We would like to acknowledge and would express our special thanks of gratitude to our honourable project guide Prof.M.V.Phatak, of Department of Computer Engineering, Maharashtra Institute of Technology, Pune for guidance and encouragement in carrying out our work. The supervision and support that she gave truly helped the progression and smoothness of the project. The co-operation is much indeed appreciated. REFERENCES [1] Studying Aesthetics in Photographic Images Using a Computational Approach Ritendra Datta, Dhiraj Joshi, Jia Li, And James Z. [2] Emotion Related Structures In Large Image Databases Martin Solli, Reiner Lenz Itn, Linkoping University Se-60174 Norrkoping, Sweden [3] Image Quality Assessment Using Multi Method Fusion Tsung-Jung Liu, Weisi Lin, C.-C.Jay Kuo 2013 Ieee Transactions On Image Pro- Cessing,Vol.22,No.5,May 2013 [4] Making A Completely Blind Image Quality Analyzer Anish Mittal,Rajiv Soundararajan And Alan C Bovik 2012ieee [5] Assessing The Aesthetic Quality Of Photographs Using Generic Image Descriptors Luca Marchesotti,Florent Perronnin,Diane Larlus,Gabriela Csurka 2011 Ieee InTernational Conference On Computer Vision. [6] Assessment Of Photo Aesthetic With Efficiency KuoYen Lo,Keng-Hao Liu And Chu-Song Chen [7] Rapid: Rating Pictorial Aesthetic Using Deep Learning Xin Lu, Zhe Lin, Jian Chaoyang, James Z Wang 2014. [8] Aesthetic Visual Quality Assessment Of Paintings Ijstsp Congcong Li,Tsuhan Chen [9] Harmony-Guided Quality Assessment Christel Chamaret And Fabrice Urban 2013 Ieee Conference On Computer Vision And Pattern Recognition Workshops [10] Content-Based Photo Quality Assessment Xiaoou Tang,Wei Luo And Xiaogang Wang 2013 [11] Content-Based Photo Quality Assessment Xiaoou Tang,Wei Luo And Xiaogang Wang 2011.
All rights reserved by www.ijsrd.com
699