Proc. of Int. Conf. on Recent Trends in Information, Telecommunication and Computing, ITC
Three-Dimensional Video Compression using Motion Estimation-Compensation and Discrete Wavelet Transform Harneet Kaur1 and Neeru Jindal2 1
Rayat Institute of Engineering and Information Technology / Electronics & Communication Engineering, Railmajra, India Email: harneet758@gmail.com 2 Rayat Institute of Engineering and Information Technology / Electronics & Communication Engineering, Railmajra, India Email: {neerusingla99@gmail.com}
Abstract— Internet data almost double every year. The need of multimedia communication is less storage space and fast transmission. So, the large volume of video data has become the reason for video compression. The aim of this paper is to achieve temporal compression for three-dimensional (3D) videos using motion estimation-compensation and wavelets. Instead of performing a two-dimensional (2D) motion search, as is common in conventional video codec’s, the use of a 3D motion search has been proposed, that is able to better exploit the temporal correlations of 3D content. This leads to more accurate motion prediction and a smaller residual. The discrete wavelet transform (DWT) compression scheme has been added for better compression ratio. The DWT has a high-energy compaction property thus greatly impacted the field of compression. The quality parameters peak signal to noise ratio (PSNR) and mean square error (MSE) have been calculated. The simulation results shows that the proposed work improves the PSNR from existing work. Index Terms—video, compression, temporal compression, DWT, PSNR, MSE.
I. INTRODUCTION A video is a sequence of frames (or images). As today is the world of telecommunication and multimedia systems, communication approach uses video to represent some information but bandwidth is still a bottleneck. Hence, video compression techniques are of prime importance for reducing the amount of information needed for picture sequence without losing much of its quality, judged by human viewers. Video contains much spatial and temporal redundancy. The goal is to efficiently exploit such types of redundancies to achieve video compression. The two-dimensional (2D) television has been quite established with digital television. The research on three-dimensional (3D) consumer electronic product has received high interest over the past decade in order to provide viewers with more realistic vision than traditional 2D video. In the area of 3D video compression, many researchers have proposed related methods, to perform 3D video and depth coding. “Reference [1]”: Zhaohui sun and Murat Tekalp; discussed about mosaic synthesis based on the trifocal motion model. “Reference [2]”: Adnan M. Alattar; developed a model for the wipe region and is used to derive the statistical characteristics of the frames in the wipe region. “Reference [3]”: Mihaelav.d. DOI: 02.ITC.2014.5.538 © Association of Computer Electronics and Electrical Engineers, 2014
Schaar-Mitrea and Peter H. N. de With; introduced various efficient algorithms that support both the lossless and the lossy compression of graphics data. “Reference [4]”: Sanghoon Lee, Marios S. Pattichis, and Alan Conrad BoviBovik; described a frame work where foveated video is created by a non uniform filtering scheme that increases the compressibility of the video stream. “Reference [5]”: Marc Alzina, WojciechSzpankowski, and AnanthGrama; proposed a lossy data compression framework based on an approximate 2D pattern matching extension of the Lempel–Ziv lossless scheme. “Reference [6]”: Laurent Itti; developed a biologically-motivated algorithm to select visually-salient regions of interest in video streams for multiply-foveated video compression. “Reference [7]”: Javier Ruiz Hidalgo, and Philippe Salembier; developed a new techniques where the compression efficiency of video codecs is improved by the use of indexing metadata. “Reference [8]”: S. Adedoyin, W.A.C. Fernando, A. Aggoun; discussed a novel approach to use both motion and disparity information to compress 3D integral video sequences. “Reference [9]”: B. Kamolrat, W. A. C. Fernando, M. Mark and A. Kondoz; addressed 3D motion estimation for depth image coding in 3D video coding. “Reference [10]”: D. V. S. X. De Silva, W. A. C. Fernando and S. L. P. Yasakethu; used object based coding of the depth maps for 3D video coding. “Reference [11]”: Ying Chen, Ye-KuiWang, Kemal Ugur, MiskaM. Hannuksela, JaniLainema, and MoncefGabbouj;worked on multiview applications and introduced solutions to support generic multiview as well as 3D services.“Reference [12]”: Anminliu, Weisilin,Manoranjanpaul,FanZhangandChenweideng; discussed a method for video in which the frames are allowed to be formed in a non-XY plane. “Reference [13]”: Qijunwang, Ruiminhu and Zhongyuanwang; described an object based depth image coding technique. “Reference [14]”: Chaminda T. E. R. Hewage, and Maria G. Martini; proposed a Reduced-Reference quality metric for color plus depth 3D video compression and transmission, using the extracted edge information of color plus depth map 3D video. “Reference [15]”: Rajeshwar Dass, Lalit Singh, and SandeepKaushik; explored the idea about different techniques available for video compression. “Reference [16]”: Rutika Joshi, Rajesh Kumar Rai, JigarRatnottar; introduced the fundamental concepts of video compression and briefly explained the various video compression algorithms and also describes the characteristics of video compression standards. The wavelet transform has greatly impacted the field of video compression. The amount of compression that can be achieved depends on the energy compaction property of the transform being used. The wavelet transform has a high-energy compaction property, progressive reconstruction that makes it a powerful tool for video compression. The proposed work discusses the 3D video compression. A motion vector can be computed using variable block size motion estimation (VBMSE). Once the motion of a macroblocks is estimated, it is then used to align with the appropriate block in the reference frame. To find the difference between the current macro-block and the aligned block in the reference frame, motion compensation is computed. This process gives motion vector in horizontal and vertical direction (x and y direction). Now estimate the motion in the 3rd dimension (also known as z direction) in the depth map to find the true motion vector. For better video frame quality, DWT compression scheme is added with motion estimation compensation video compression technique. As a result proposed work shows an improvement in terms of peak signal to noise ratio (PSNR) and compression ratio (CR). The paper is organized as follows: Including Sect. I, Sect. II describes the fundamental concepts of 3D motion estimation- compensation. This is followed by the proposed work in Sect. III. Simulation results are discussed in Sect. IV. Sect. V includes conclusion and future scope. II. 3D MOTION ESTIMATION-COMPENSATION The motion estimation plays a vital role in video coding by taking an advantage of temporal redundancy across the sequence. The differences between two consecutive frames are mainly due to movement of objects. Using a model of the motion of the objects between frames, the encoder estimates the motion that occurred between the reference frame and the current frame. The output of this process will be motion estimation. “Fig. 1 (a) (b)” shows the reference frame and the current frame respectively. These are the two consecutive frames taken as a reference from video sequence. “Fig. 1 (c)” shows the reference frame, where measurement window is compared with shifted array of pixel, to determine best match and “Fig. 1 (d)” shows current frame after segmented into macroblocks. The important goals of motion estimation method are to increase the prediction accuracy, or to reduce the implementation complexity, or both. The translational motion estimation of the block of pixels in the current frame with respect to the previous frame can be accomplished by block-matching algorithm. To determine the motion of an object in a frame, the first step is to segment the frame into macroblocks of different sizes and then define the object boundaries
208
(a)
(b)
(c)
(d)
Figure 1. Shows concept of motion estimation (a) Reference frame and (b) current frame (c) current frame segmented into macroblocks and (d) reference frame with motion vectors
in the current and the reference frames. “Fig. 2 (a)” shows frame 1 of “Interview” test sequence and “Fig. 2 (b)” shows frame with 8 × 8 pixel block size distribution.
(a)
(b)
Figure 2. Shows (a) Frame 1 (b) 720 × 576 frame is segmented into macroblocks each of size 8 pixel by 8 pixel of “Interview” test sequence
To find the best match search an area in the reference frame. This is carried out by comparing the M × N block in the current frame with the entire possible M × N block in the search area in the reference frame. This will yield a motion vector with components in the horizontal and vertical directions. The “Fig. 3 (a) (b)” describes the motion estimation in x and y direction. 3D depth map search is explained in “Fig. 3 (c)”. Once the motion vector is determined, for calculating motion compensation the rectangular block in the current frame can be aligned with that in the reference frame and the corresponding differential pixels can be found. The performance for 2D motion estimation is calculated with mean square error (MSE) as in “(1)”. zmin
Current block M N
(a)
(a+i,b+j) (a,b)
(a+i,b+j)
zmax
(a,b)
(b)
(c)
Figure 3. Shows block matching for motion estimation (a) current block in reference frame (b) block displaced in x and y direction in current frame (c) 3D motion estimation
209
MSE ( x, y ) =
1 M×N
M
N
∑ ∑ ( f (m , n ) − g (m + x, n + y ))
2
(1)
m =1 n =1
where f(m,n) represents the current block of M×N pixels at coordinates (m,n) and g(m+x,n+y) represents the corresponding block in the reference frame at new coordinates. Examples of per pixel depth map frame of the “Interview” test sequence are shown in “Fig. 4”.
(a)
(b)
Figure 4. Shows (a) Frame 25 of “Interview” test sequence (b) with its associated depth map
In addition to the motion search in x and y direction, estimating the motion in the z direction involves incrementing or decrementing the luminance values of all the pixels within a macroblock, until a pre-defined error metric is minimized. This is given mathematically as in “(2)”. MSE ( z ) =
1 M ×N
M
N
∑ ∑ ( f (m , n ) − g (m + x, n + y ) + z )
2
( 2)
m =1 n =1
where the searching window in the depth dimension is defined as zmin<zi<zmax. In case of 8 bits per pixel, the depth value can be varied between 0-255. Therefore a motion vector in z component can be an integer value in the range of -255 to +255. III. PROPOSED WORK The objective of the proposed technique is to compress the 3D video using block-matching approach. In this approach, the two consecutive frames i.e. the previous and the current frame respectively are taken. The current frame is divided into non-overlapping macroblocks of different sizes and for each candidate block, the best motion vector is determined (in the reference frame). Thus the interframe redundancy is removed. The 2D block matching is performed in x and y direction. A motion vector evaluated during the procedure describes the offset between the location of the block being coded in the current frame and the location of the best matching block in the reference frame and is recorded. The motion compensated prediction frame is then formed from the entire shifted region from the entire previous decoded frame. Then using the 2D motion information, block matching is performed in the depth direction to find the z direction motion vector components. This minimizes the motion compensation error. Finally, the predicted motion compensated frame will be the input to DWT compression scheme to get the better PSNR with better video quality frame. The flow chart of proposed work is shown in “Fig. 5”. Starting from extracting the frame from the video sequence, next, the frames are segmented into macroblocks of different size (16 pixel by 16 pixel, 8 pixel by 8 pixel and 4 pixel by 4 pixel) for estimating motion vectors in x and y components. A full search strategy is used while evaluating the results. It computes the appropriate matching block within the search window for all the possible integer-pel displacements as explained in “Fig. 6 (a)”. For reference, the motion estimation of a macroblock involves finding the 16 × 16 pixel block in the search area in a reference frame that closely matches the current macroblock as shown in “Fig. 6 (b)”. The best matching block is found in the reference frame and it becomes the predictor for the current M × N block and is subtracted to form a residual M × N block. The residual block is encoded and transmitted and the offset between the current block and the position of the block in the reference frame i.e. the motion vectors computed is also transmitted. The candidate block that minimizes the residual energy is chosen as the best match. 210
Start Input video
Extracting frame from video
16×16 Block search
8×8 Block search
Calculate SAD1
Calculate SAD 2
4×4 Block search Calculate SAD 3
IS SAD 1 < T SAD 2 < T SAD 3 < T
No
Yes
Use block size with minimum SAD value Depth direction search DWT compression scheme Check the PSNR value Stop Figure 5. Flowchart of proposed work
The best match block is checked with sum of absolute difference (SAD) as matching criteria. It makes the error value positive, instead of summing up the squared difference, the absolute difference are summed up. The SAD measured at displacement (i,j) is computed as shown in “(3)”. M
N
SAD (i, j ) = ∑ ∑ s( x + m, y + n ) − s′( x + m + i , y + n + j ) (3) m =1 n =1
where, (x,y) is the position of current macro-block, (i,j) is the position of reference macroblock, s is current macroblock pixel value, s′ is the pixel value of macroblock in a reference frame and M×N is the size of macroblock. A threshold (t) value is set before, and then SAD is evaluated as a matching criterion with compression ratio as in “(4)”. Compression Ratio =
Compressed Size Uncompressed Size
(4)
where, uncompressed size is the original size of frame of video sequence taken and compressed size is the size of reconstructed frame (or compressed frame). The minimum SAD value, block size is selected.
211
Raster Scan Initial search location
Center position
(a)
Reference frame Search area Best match
(b) Figure 6. (a) Full search (b) block-matching in the reference frame
Using the 2D motion information, suitable depth map motion vector are calculated by adjusting the depth parameter during the procedure as shown in â&#x20AC;&#x153;Fig. 7â&#x20AC;?.
Figure 7. 3D motion information (a) Original depth map (b) reference frame at increase depth map (c) reference frame at decrease depth map
212
To improve the accuracy of motion prediction, the z direction motion (3D motion) to search for the depth map z-zi (decreases depth map) and z+zi (increase depth map) has been adopted. This approach provides more accurate information in 3D and better exploits the temporal correlations of 3D content. Thus video sequence can be compressed easily by using motion information of x, y and z direction. The DWT enables the computer to store an image in many scales and resolutions. Hence it allows one to compress a frame using less storage space with more detail. Thus, compression using 3D information can be further improved by using DWT compression scheme. To compare the performance of video sequence compressed after 3D motion estimation and video sequence compressed after applying wavelet compression scheme, PSNR has been computed using “(5)”. PSNR = 10 log10
MAX 2 MSE ( f , g )
(5)
where, MAX=255, ‘f’ is the original size of video frame taken, and ‘g’ is the restored size of video frame taken. It is measured in decibels (dB). For a given sequence, high PSNR usually indicates high quality and low PSNR usually indicates low quality. IV. SIMULATION RESULTS AND DISCUSSION The performance of the proposed work has been evaluated on various video sequences i.e. “Interview”, “Kendo”, “Balloons”, “Gstennis” and “Inition-2d-3d-Showreel” available from the Tanimoto Laboratory [17]. For reference, the luminance components of the first frame of test sequences are shown in “Fig. 8”.
(a)
(b)
(d)
(c)
(e)
Figure 8. Frame 1 of (a) “Kendo” (b) “Balloons” (c) “Gstennis” (d) “Inition-2d-3d-Showreel” and (e) “Interview” video test sequence
The resolution of all the test sequence used is 720 × 576 pixel at different frame rates. Using block based motion estimation, the resulting motion vector is easy to represent by one motion vector per block, and thus it is achieves good and robust performance for compression. For each sequence PSNR, MSE and CR have been computed for frame 1 to frame 9 for comparing the performance. In this simulation, results are shown using block size 4 × 4 and window size 2N × 2N, where N is block size. Specifically, if objects within the scene have fast motion, large window size is desirable and small window size is preferred for sequences with slow motion. The PSNR computed at CR equal to 51.13% is shown in Table I. The average value of quality parameter PSNR of test frame “Kendo” and “Balloons” has been improved from 32.88 dB to 41.06 dB and 41.72 dB to 45.44 dB respectively. From the results evaluated 213
in Table I, it has been observed that the average value of test frame “Gstennis” is 51.11 dB and test frame “Inition-2d-3d-Showreel” is 40.57 dB. Simulation results of “Interview” test sequence i.e. the reconstructed frame and compressed frame using proposed approach are shown in “Fig. 9 (a)” and “Fig. 9 (b)” respectively. TABLE I. PSNR OF VARIOUS TEST SEQUENCE WITH PROPOSED METHOD Frame PSNR(dB) of “Kendo” test PSNR(dB) of “Balloons” test PSNR(dB) of “Gstennis” test sequence sequence sequence No.
PSNR(dB) of “Inition-2d-3dShowreel” test sequence
P1
40.81
45.55
52.11
40.57
P2
41.43
45.55
52.25
40.58
P3
40.91
45.48
51.43
40.58
P4
40.96
45.42
50.88
40.54
P5
41.04
45.49
50.95
40.52
P6
41.05
45.44
50.99
40.54
P7
41.10
45.37
50.60
40.63
P8
41.13
45.31
50.44
40.60
P9
41.13
45.35
50.36
40.60
(a)
(b)
Figure 9. “Interview” test sequence (a) reconstructed frame (b) compressed frame using proposed approach
The most video compression systems are designed to minimize the MSE between two video sequences. It is an objective quality matrix commonly used in video compression to calculate an error due to their simplicity and ease of calculations. The MSE evaluated for various video test sequences using proposed approach is shown in Table II. TABLE II. MSE OF VARIOUS T EST SEQUENCE WITH PROPOSED METHOD Frame MSE of “Kendo” test MSE of “Interview” test MSE of “Balloons” test MSE of “Gstennis” test sequence sequence sequence sequence No.
MSE of “Inition-2d-3dShowreel” test sequence
P1
5.39
72.03
1.81
1.39
5.70
P2
4.67
73.73
1.81
1.38
5.67
P3
5.26
72.00
1.83
1.46
5.68
P4
5.20
73.77
1.86
1.53
5.73
P5
5.11
72.07
1.83
1.52
5.76
P6
5.09
73.86
1.85
1.51
5.73
P7
5.03
72.36
1.88
1.56
5.61
P8
5.00
74.05
1.91
1.58
5.65
P9
5.00
72.39
1.89
1.59
5.66
214
The 3D depth map search is performed and depth map parameter is adjusted during the procedure by varying each pixel value between zmin to zmax. Finally, the value between 0-255 which minimizes the motion compensation error is selected as the motion vector in the z component. The depth map motion compensation using 3D motion vectors of “Interview” test sequence is shown in “Fig. 10 (a)”. “Fig. 10 (b)” shows depth map motion compensation using 3D motion vectors and wavelets.
(a)
(b)
Figure 10. “Interview” test sequence (a) depth map motion compensation using 3D motion (b) depth map motion compensation using wavelets compression scheme
The advantage of decomposing frame to approximate and detail parts as it enables to isolate and manipulate the data with specific properties. With this, it is possible to determine whether to preserve more specific details. This would allow the frame to lose a certain amount of horizontal and diagonal details, but would not affect the frame in human perception. As from the results shown in “Fig. 10” it can be observed that, by using proposed approach frame can be compressed with highly good picture quality. The results obtained with the proposed method are compared against variable block size motion estimation. The PSNR can be calculated easily and quickly and is therefore a very popular quality measure, widely used to compare the ‘quality’ of compressed and decompressed video frames. The results obtained with proposed method shows better in terms of PSNR as shown in Table III. TABLE III. C OMPARISON OF VARIABLE B LOCK SIZE MOTION ESTIMATION WITH PROPOSED M ETHOD OF “INTERVIEW ” TEST SEQUENCE Frame No. PSNR(dB) of VBMSE PSNR(dB) of Proposed methodImprovement (dB) P1
26.06
29.55
3.49
P2
25.90
29.45
3.55
P3
26.08
29.55
3.47
P4
26.01
29.45
3.44
P5
26.21
29.55
3.34
P6
25.99
29.44
3.45
P7
25.86
29.53
3.67
P8
25.92
29.43
3.51
P9
25.80
29.53
3.73
The proposed method works well for the “Interview” sequence, which is a sequence with low motion. From the Table III it has been observed that performance of quality metric PSNR is improved by 3.51 dB averagely with proposed method at high compression ratio. The simulation condition of test frame “Interview” is under compression ratio equal to 51.29%, test frame “Kendo” is 52.50%, test frame “Balloons” is 51.33%, test frame “Gstennis” is 51.37% and test frame “Inition-2d-3d-Showreel” is 51.46%. The “Balloons” test sequence is a high motion sequence; and the simulation results evaluated on this sequence are shown in “Fig. 11”. The reference frame, current frame, its 8 × 8 block size distribution can be 215
seen in “Fig. 11 (a), (b) and (c)” respectively. The reconstructed frame i.e. the compressed frame after applying VBMSE algorithm is shown in “Fig. 11 (d)”. The compressed frame with proposed method is shown in “Fig. 11 (e)”.
(a)
(b)
(c)
(d)
(e) Figure 11. Shows (a) Reference or previous frame (frame 1) (b) current frame (frame 2) (c) 4 × 4 block size distribution (d) reconstructed frame and (e) shows compressed frame after DWT compression scheme of “Balloons” test sequence with size 720 × 576 pixel
From “Fig. 11” it has been observed that, the frame with proposed method is compressed with good picture quality at high compression ratio. The results obtained with proposed method on “Gstennis” test sequence are shown in “Fig. 12”. It is a sequence with slow motion. By using VBMSE algorithm, the frame is compressed at good PSNR. With the proposed method furthermore, the video sequence is compressed at high compression ratio. The previous frame and current frame are shown in “Fig. 12 (a)” and “Fig. 12 (b)” respectively. During the procedure, it has been observed that value of SAD is minimum with block size 4 × 4 pixels as it provides more precise estimation. A smaller block sizes can produce better motion compensation results. However, a smaller block size leads to increased complexity and an increase in the number of motion vectors that need to be transmitted. The distribution of frame with block size 4 × 4 is shown in “Fig. 12 (c)”. “Fig. 12 (d)” shows the reconstructed frame after VBMSE algorithm and compressed frame with proposed method is shown in “Fig. 12 (e)”.
216
(a)
(b)
(c)
(d)
(e) Figure 12. Shows (a) Reference frame or previous (frame 1) (b) current frame (frame 2) (c) 4 × 4 block size distribution of frame (d) reconstructed frame and (e) shows the compressed frame after DWT compression scheme of “Gstennis” test sequence with size 720 × 576 pixel
From the results shown in “Fig. 12” it can be clearly observed that compressed frame after applying the wavelet compression scheme is compressed with better picture quality at high resolution. Next, the results evaluated on “Inition-2d-3d- Showreel” test sequence are shown in “Fig. 13”. This is a sequence with complex motion and in this type of objects, motion is hard to compensate. The different motion compensation block sizes produce different motion compensation results. The previous frame and the current frame are shown in “Fig. 13 (a)” and “Fig. 13 (b)” respectively. The evaluation of proposed method has been performed on different block size using same searching technique. The distribution of the current frame with block size 4 × 4 is shown in “Fig. 13 (c)”. The result shows that the proposed method with block size 4 × 4 pixel provides the best performance. “Fig. 13 (d)” shows the reconstructed frame obtained using VBMSE algorithm. After VBMSE algorithm, the compressed frame with proposed method is shown in “Fig. 13 (e)” and it can be observed that the proposed method fits well. The quality of compressed frame is good at high compression ratio. Overall, the proposed method provides improved results in terms of PSNR at better compression ratio.
217
(a)
(b)
(c)
(d)
(e) Figure 13. Shows (a) Reference or previous frame (frame 1) (b) current frame (frame 2) (c) 4 × 4 block size distribution (d) reconstructed frame and (e) shows the compressed frame after DWT compression scheme of “Inition-2d-3d- Showreel” test sequence with size 720 × 576 pixel
V. CONCLUSIONS The motion estimation and compensation with block matching criteria is a technique for video compression. It is preferred due to its simplicity and good compromise between prediction qualities. Motion overhead wavelet based coding provides substantial improvement in picture quality at high compression ratio mainly because of better energy compaction property of wavelet transforms. Thus after adding DWT compression technique along with depth map motion estimation and compensation for 3D video compression, an improved performance in terms of PSNR and CR is obtained with a high picture quality. For future work, several improvements can be made along with motion estimation-compensation and wavelets for three dimensional videos, another compression technique should be added to improve the quality parameter PSNR. To further improve the compression performance, another searching technique instead of full search can be implemented. Instead of wavelets another compression technique to compress video can also be implemented. Scanning can be changed to increase the consecutive number of zeros. Even several scanning can be determined before and decided at the time of encoding.
218
ACKNOWLEDGEMENTS Foremost, I would also like to thank God for the strength that keep me standing and for the hope that keep me believing that this report would be possible. I would also like to thank authors whose works I have consulted and quoted in this work. I would like to express my sincere gratitude to Neeru Jindal for her valuable guidance and full support in carrying out this work. At last, but not the least my gratitude towards my parents and friends for support and encouraging me with their best wishes. REFERENCES [1] Zhaohui Sun and A. Murat Tekalp “Trifocal Motion Modeling for Object-Based Video Compression and Manipulation,” IEEE transactions on circuits and systems for Video Technology, Vol. 8, No. 5, pp. 667-685, Sept. 1998. [2] Adnan M. Alattar, Alattar “ Wipe Scene Change Detector for use with Video Compression Algorithms and MPEG7,” IEEE Transactions on Consumer Electronics, Vol. 44,No. 1, pp. 43-51, Feb.1998. [3] Mihaelav.d. Schaar-Mitrea’ and Peter H. N. de With “Hybrid Compression of Video with Graphics in DTV Communication Systems, “IEEE Transactions on Consumer Electronics, Vol. 46, No. 4, pp. 1007-1017, Nov. 2000. [4] Sanghoon Lee, Marios S. Pattichis, and Alan Conrad Bovik “Foveated Video Compression with Optimal Rate Control,” IEEE Transactions on image processing, Vol. 10, No. 7, pp 977-992, July 2001. [5] Marc Alzina, WojciechSzpankowski and AnanthGrama “2D-Pattern Matching Image and Video Compression: Theory, Algorithms and Experiments,” IEEE Transactions on Image Processing, Vol. 11, No. 3, pp.318-331, March 2002. [6] Laurent Itti “Automatic Foveation for Video Compression using a Neurobiological Model of Visual Attention,” IEEE Transactions on Image Processing, Vol. 13, No. 10, pp.1304-1318, Oct. 2004. [7] Javier Ruiz Hidalgo and Philippe Salembier “On the Use of Indexing Metadata to Improve the Efficiency of Video Compression,” IEEE Transactions on Circuits and Systems for Video Technology, Vol. 16, No. 3, pp.410-419, March 2006. [8] S. Adedoyin, W.A.C. Fernando, and A.Aggoun “A Joint Motion & Disparity Motion Estimation Technique for 3D Integral Video Compression using Evolutionary Strategy,” IEEE Transactions on Consumer Electronics, Vol. 53, No. 2, pp.732-739, May 2007. [9] B. Kamolrat, W. A. C. Fernando, M. Mark and A.Kondoz “3D Motion Estimation for Depth Image Coding in 3D Video Coding,” IEEE Transaction on Consumer Electronics, Vol. 55, pp. 824–830, May 2009. [10] D. V. S. X. De Silva, W. A. C. Fernando and S. L. P. Yasakethu “Object Based Coding of the Depth Maps for 3D Video Coding,” IEEE Transaction on Consumer Electronics, Vol. 55, pp.1699–1706, Aug. 2009. [11] Ying Chen, Ye-KuiWang, Kemal Ugur,MiskaM. Hannuksela,JaniLainema, and MoncefGabbouj “The Emerging MVC Standard for 3D Video Services,” EURASIP Journal on Advances in Signal Processing, Volume 2009, Article ID 786015, 13 pages. [12] Anminliu, Weisilin, Manoranjanpaul, Fan ZhangandChenweideng “Optimal Compression Plane for Efficient Video Coding,”IEEE Transactions on Image Processing, Vol. 20, No. 10, pp. 2788-2798, Oct. 2011. [13] Qijunwang, Ruiminhu and Zhongyuanwang “Intracoding and Refresh with Compression-oriented Video Epitomic Priors, “ IEEE Transactions on Circuits and Systems for Video Technology, Vol. 22, No. 5, pp.714-726, May 2012. [14] Chaminda T. E. R. Hewage, and Maria G. Martini “Edge-Based Reduced-Reference Quality Metric for 3-D Video Compression and Transmission,” IEEE Journal of selected topics in signal processing, Vol. 6, No. 5, pp. 471-482, Sep. 2012. [15] RajeshwarDass , Lalit Singh, SandeepKaushik “Video Compression Technique,” International Journal of Scientific & Technology Research, Vol. 1, Issue 10, pp. 114-119, Nov. 2012. [16] Rutika Joshi, Rajesh Kumar Rai, JigarRatnottar “Review of Different Standards for Digital Video Compression Technique,” International Journal of advancement in electronics and computer engineering (IJAECE), Vol. 1, Issue 1, pp.32-38, April 2012. [17] TanimotoLaborotory, Nagoya University, “http://www.tanimoto.nuee.nagoya-u.ac.jp/”.
219