24
COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS Submitted by, Ashwini Urs and Sharath Patil Under guidance of Dr.K.R.Rao

COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Embed Size (px)

DESCRIPTION

COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS. Submitted by, Ashwini Urs and Sharath Patil Under guidance of Dr.K.R.Rao. Introduction. Integer DCT. KLT is the statistically optimal transform. The performance of DCT is close to the performance of KLT [1]. - PowerPoint PPT Presentation

Citation preview

Page 1: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Submitted by,

Ashwini Urs and Sharath Patil

Under guidance of

Dr.K.R.Rao

Page 2: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Introduction

Page 3: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Integer DCT

• KLT is the statistically optimal transform.• The performance of DCT is close to the

performance of KLT [1].• DCT is a well-known transform and is widely

used by majority of coding standards.• Though integer DCT contains only integers, it

has similar energy-packing ability as that of DCT [1].

Page 4: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Integer DCT (Continued)

• Integer cosine transform does not involve floating point computations and hence is used in video coding standards such as H.264 [2], VC-1 [3] and AVS [4].

• Integer cosine transform has been implemented with transform sizes of 4, 8 and 16 [1].

• Even larger size transforms (up to 64) have been used for high resolution videos to achieve higher coding gain [1].

Page 5: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Integer DCTs compared

Page 6: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Integer DCT matrix for AVS-China, H.264 and VC-1

AVS-China [2]

2691010962

410104410104

6102992106

88888888

9210661029

104410104410

1096226910

88888888

361012121063

48844884

612310103126

88888888

103126612310

84488448

121063361012

88888888

H.264 [3]

491516161594

616166616166

916415154169

1212121212121212

154169916415

166616166616

161594491516

1212121212121212

VC-1 [4]

Page 7: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Integer DCT matrix for AVS-China, H.264 and VC-1

• The orthogonality of the 3 matrices was checked by evaluating [INTDCTi] x [INTDCTi]*T

.

• The orthogonalised matrices are:1. AVS-China = diag (512, 442, 464, 442, 512, 442,

464, 442)2. H.264 = diag (512, 578, 320, 578, 512, 578, 320,

578)3. VC-1 = diag (1152, 1156, 1168, 1156, 1152, 1156,

1168, 1156)

Page 8: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Order-16 Integer DCT matrix used in AVS-China [26]

8888888888888888

10109966222266991010

101044441010101044441010

99221010666610102299

8888888888888888

441010101044441010101066

441010101044441010101044

22669910101010996622

22669910101010996622

441010101044441010101044

66101022999922101066

8888888888888888

99221010666610102299

101044441010101044441010

10109966222266991010

8888888888888888

16T

Page 9: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Comparison of the properties of integer DCTs

Page 10: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Comparison of interger DCT matrices

• The properties of the 3 integer DCT matrices were compared by considering a covariance matrix R for a Markov-I process with ρ = 0.95 and N=8.

• Rjk = [ρ|j-k|] for j, k = 0, 1,…, N-1, where ρ is the adjacent correlation coefficient.

• Covariance matrix in transform domain is given by

where DOT is discrete orthogonal transform and [Σ] is the covariance matrix in spatial

*~

TDOTDOT

Page 11: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Properties used for comparison of integer DCTs

1. Variance distribution: The diagonal elements of correspond to the variances in the transform domain [7].

2. Rate versus distortion: RD is the minimum average rate (bits/sample) for coding a signal at a specified distortion D [7]. For fixed average distortion D, rate distortion function RD is computed as

Choose values of θ betweent 0.1 and 1. For the same values of θ, D and RD are calculated [7].

Page 12: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Properties used for comparison of integer DCTs

3. Normalized basis restriction error, Jm: The compaction of energy in a few transform coefficients can be represented by the normalized basis restriction error defined as [7]:

where are arranged in decreasing order [7].2~

kk

Page 13: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Properties used for comparison of integer DCTs

4. Residual correlation: An indication of the extent of decorrelation in transform domain can be gauged by correlation left undone by the discrete transform, which is measured by the absolute sum of cross-covariance (off diagonal elements) in the transform domain i.e.,

for N = 8 as a function of ρ [7].

Page 14: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Properties used for comparison of integer DCTs

5. Transform coding gain GTC: Transform coding gain is defined as the ratio of arithmetic mean to geometric mean of variances

where is the variance of the ith co-efficient in the transform domain.

• As sum of all the variances is in invariant under orthogonal transformation, by minimizing geometric mean GTC can be maximized [7].

2~

ii

2~

ii

Page 15: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Results and Conclusion

Page 16: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Variance distribution versus N

Page 17: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Rate versus distortion

Page 18: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Normalized basis restriction error versus samples retained m

Page 19: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Residual correlation versus correlation co-efficient

Page 20: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

Conclusion

• Variance distribution, normalized basis restriction error and transform coding gain of these 3 codecs are almost comparable.

• Transform coding gain, GTC for AVS, H.264 and VC-1 are 8.2916, 8.0155 and 7.5477 respectively. From this, we observe that AVS achieves maximum GTC.

• For a fixed average distortion D, the rate distortion function characteristics of H.264 and AVS are indistinguishable.

• The residual correlation for ρ > 0.5 is indistinguishable for these 3 codecs.

Page 21: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

References[1] C. Fong and W. Cham, “Simple order-16 integer transform for video coding”, The Chinese university of Hong Kong, Shatin, Hong Kong.

[2] S.K.Kwon, A.Tamhankar and K.R.Rao, “Overview of H.264 / MPEG-4 Part 10” J. Visual Communication and Image Representation, vol. 17, pp.186-216, April 2006.

[3] S. Srinivasan , et al, “Windows Media Video 9: overview and applications”, Signal Processing: Image Communication, vol. 19, Issue 9, pp. 851-875, Oct. 2004

[4] W. Gao et al., “AVS – The Chinese next-generation video coding standard,” National association of broadcasters, Las Vegas, 2004

[5] R. Joshi, Y. Reznik and M. Karczewicz, “Efficient large size transforms for high-performance video coding”, Qualcomm Inc., San Diego, CA, USA.

[6] “Integer DCT for AVS China”, INTDCT6 - http://www-ee.uta.edu/dip/Courses/EE5355/ee5355.htm.

Page 22: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

References[7] “Comparison of discrete transforms”, http://www-ee.uta.edu/dip/Courses/EE5355/ee5355.htm.

[8] N.Ahmed, T.Natarajan and K.R.Rao, “Discrete cosine transform”, IEEE trans. computers, Vol. X, pp.90-93, 1974.

[9] A.K.Jain, “Fundamentals of digital image processing”, Prentice hall, 1989.

[10] A.T. Hinds, “Design of high-performance fixed-point transforms using the common factor method”, Ricoh I infoprint solutions company, Boulder, CO, USA.

[11] T.Wiegand, et al “Overview of the H.264/AVC video coding standard”, IEEE Trans. on Circuit and Systems for Video Technology, vol.13, pp. 560-576, July 2003.

[12] T. Wiegand and G. J. Sullivan, “The H.264 video coding standard”, IEEE Signal Processing Magazine, vol. 24, pp. 148-153, March 2007.

Page 23: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

References[13] D. Marpe, T. Wiegand and G. J. Sullivan, “The H.264/MPEG-4 AVC standard and its applications”, IEEE Communications Magazine, vol. 44, pp. 134-143, Aug. 2006.

[14] A. Puri, X. Chen and A. Luthra, “Video coding using the H.264/MPEG-4 AVC compression standard”, Signal processing: image communication, vol. 19, pp. 793-849, Oct. 2004.

[15] M.Fieldler, “Implementation of basic H.264/AVC decoder”, seminar paper at Chemnitz university of technology, June 2004.

[16 ]R. Schäfer, T. Wiegand and H. Schwarz, “The emerging H.264/AVC standard”, EBU Technical Review, Jan. 2003.

[17]D. Marpe, T. Wiegand, and S. Gordon, "H.264/MPEG4-avc fidelity range extensions: tools, profiles, performance, and application areas," in, IEEE international conference on image processing, vol. 1, pp. I-593-6, 2005.

[18] S. Saponara et al, "The JVT advanced video coding standard: complexity and performance analysis on a tool-by-tool basis," in Packet Video Workshop, Nantes, France, April 2003.

[19] VC-1 technical overview - http://www.microsoft.com/windows/windowsmedia/howto/articles/vc1techoverview.aspx

Page 24: COMPARISON OF 8 × 8 INTEGER DCTs USED IN H.264, AVS-CHINA AND VC-1 VIDEO CODECS

References[20] S. Srinivasan and S. L. Regunathan, “An overview of VC-1”, SPIE / VCIP, vol. 5960, pp. 720-728, July 2005.

[21] AVS Video Expert Group, “Information technology – Advanced coding of audio and video – Part 2: Video (AVS1-P2 JQP FCD 1.0),” Audio Video Coding Standard Group of China (AVS), Doc. AVS-N1538, Sept. 2008.

[22] AVS Video Expert Group, “Information technology – Advanced coding of audio and video – Part 3: Audio,” Audio Video Coding Standard Group of China (AVS), Doc. AVS-N1551, Sept. 2008.

[23] L Yu et al., “Overview of AVS-Video: Tools, performance and complexity,” SPIE VCIP, vol. 5960, pp. 596021-1~ 596021-12, Beijing, China, July 2005. [24] L. Fan, S. Ma and F. Wu, “Overview of AVS video standard,” IEEE Int’l Conf. on Multimedia and Expo, ICME '04, vol. 1, pp. 423–426, Taipei, Taiwan, June 2004. . [25] Special issue on 'AVS and its Applications' Signal processing: image communication, vol. 24, pp. 245-344, April 2009.

[26] C. K. Fong and W. K. Cham, “Simple order-16 integer transform for video coding”, http://www-ee.uta.edu/Dip/Courses/EE5355/INTDCT5.pdf