Brief Overview of Wyner-Ziv CODEC and Research Plan Jin-soo KIM Contents Overview of Wyner-Ziv CODEC Application of Wyner-Ziv CODEC Basic Principle of WZ CODEC Generation of S.I. at the Decoder How to Encode WZ frames Research Plan Q&A 2 J.S.Kim Application of WZ CODEC 3 J.S.Kim Video coding : history and trends 2010 2005 H.265(?) HDTV 2003 H.264 SVC mobile Mobile TV 1999 MPEG4 Hand PC Video Conferencing H.263 MPEG2 Mobile Phone 1994 1992 MPEG1 Mobile : 3low1high Year • Low (battery, bandwidth, CPU) • High cost 4 J.S.Kim (Conventional) Interframe Video Coding Predictive Interframe Encoder X Predictive Interframe Decoder Y Y X’ Side Information 5 J.S.Kim Low Complexity Encoder Wyner-Ziv Intraframe Encoder Wyner-Ziv Interframe Decoder X Y X’ Side Information [Witsenhausen, Wyner, 1980] [Puri, Ramchandran, Allerton 2002] [Aaron, Zhang, Girod, Asilomar 2002] … 6 J.S.Kim Applications of WZ codec Light encoder and light decoder B. Girod, A. Aaron, S. Rane, D. Rebollo-Monedero, “Distributed video coding,” Proceedings of the IEEE, Vol93, pp71-83, Jan. 2005. 7 J.S.Kim Applications of WZ codec http://www.discoverdvc.org/deliverables/Discover-D4.pdf Wireless low power video surveillance Disposable video cameras Sensor network Multi-view image acquisition Medical applications Networked camcoders 8 J.S.Kim Applications of WZ codec http://www.discoverdvc.org/deliverables/Discover-D4.pdf SensorCamPillCamWearableCamDispos able cam.ScanCam 9 J.S.Kim Basic Principle of WZ CODEC 10 J.S.Kim Lossless Compression with Side Information R ≥ H(X|Y) X Statistically dependent Encoder Y Decoder X Y Side Information Wyner-Ziv showed that the conditional rate-mean squared error distortion function for X is the same whether the side information Y is available only at the decoder, or both at the encoder and the decoder. R ≥ H(X|Y) X Encoder Decoder Statistically dependent Side Information 11 Y X [Slepian, Wolf, 1973] J.S.Kim Shannon Theory with side info. Example) x : dice number H(X) = 6Σlog26 = 2.58 bits Shannon coding theorem X encoder R decoder X Y No error, if H(X) < R(X) = 3 bits If R(X) = 2, {00,01,10,11}{1,2,{3,4},{5,6}} Information loss With side information Y=“even number” H(X|Y) = 3Σlog23 = 1.58 < R(X|Y) = 2 12 J.S.Kim Wyner-Ziv coding (lossy) A. Majumdar, R. Puri, P. Ishwar, K. Ramchandran, “Complexity/performance trade-offs for robust distributed video coding,” IEEE ICIP2005, Vol. 2, pp678-81, 11-14 Sept. 2005 • • • WZ = quantization + Slepian-Wolf Random coset partitioning operation, 3bit-info can be represented by 2bit (LSB first increase Δ) • X : original value U : quantized value • Y : side information in the decoder • given Y + sent 10U=101 13 J.S.Kim History of DVC Slepian and Wolf : lossless DVC (1973) Wyner and Ziv : lossy DVC (1976) Power-efficient, Robust, hIgh-compression, Syndrome-based Multimedia coding Girod in Stanford : Good review (2005) “The rate-distortion function for source coding with side information at the decoder,” IEEE Tr. Information Theory, 1976. Ramchandran in Berkeley : PRISM (2002) “Noiseless coding of correlated information sources,” IEEE Tr. On Information Theory, 1973. “Distributed video coding,” IEEE Proceedings, 2005. EU : DISCOVER(~2006), www.discoverdvc.org DIStributed COding for Video sERvices 14 J.S.Kim Towards Practical Slepian-Wolf Coding Convolution coding for data compression [Blizard, 1969] Convolutional source coding [Hellman, 1975] Syndrome source coding [Ancheta, 1976] Coset codes [Pradhan and Ramchandran, 1999] Trellis codes [Wang and Orchard, 2001] Turbo codes [García-Frías and Zhao, 2001] [Bajcsy and Mitran, 2001] [Aaron and Girod, 2002] LDPC codes [Liveris, Xiong, and Georghiades, 2002] ... ... 15 J.S.Kim Generation of S.I. at the Decoder 16 J.S.Kim Motion Compensation Motion-compensated interpolation (MC-I) using the decoded Key frame at time t-1 & t+1 Estimate the motion t-1 Key Frame t t+1 Wyner-Ziv Frame Key Frame 17 J.S.Kim Side Information Key Frame Conventional Intraframe codec K1 K1' Reconstructed Key Frame Side Information Wyner-Ziv Frame S2 Interpolation S2 Side Information Wyner-Ziv Frame S3 Interpolation S3 S3' Reconstructed WZ Frame Side Information Wyner-Ziv Frame S4 Key Frame K5 Wyner-Ziv Decoder Conventional Intraframe codec 18 Interpolation K5' S4 Reconstructed Key Frame J.S.Kim Motion Compensation Motion-compensated extrapolation (MC-E) estimate the motion between the Wyner-ziv frame a t time t-2 and the Key frame at time t-1 Estimate the motion t-2 t-1 t Wyner-Ziv Frame Key Frame Wyner-Ziv Frame 19 J.S.Kim Side Information Key Frame K1 Conventional Intraframe codec K1' Reconstructed Key Frame S2' Reconstructed WZ Frame Side Information Wyner-Ziv Frame S2 Extrapolation S2 Wyner-Ziv Decoder Wyner-Ziv Frame S3 Extrapolation S3 S3' Reconstructed WZ Frame Wyner-Ziv Decoder 20 J.S.Kim Motion Compensation 1 WZ Frame : K1 - S2 - K 3 3 WZ Frame : K1 - S2 - S3 - S4 - K 5 1 S 2 MCI ( K , K , ) 2 1 S 3 MCI ( K , K , ) 2 ^ ' ' 1 S 2 MCI ( K1 , S3 , ) 2 ^ ' ' 1 S 4 MCI ( S 3 , K 5 , ) 2 ^ ' 1 ' 3 2 WZ Frame : K1 - S2 - S3 - K 4 1 S 2 MCI ( K1' , K 4' , ) 3 ^ 2 S 3 MCI ( K1' , K 4' , ) 3 ^ ^ 21 ' 1 ' 5 J.S.Kim How to Encode WZ frames 22 J.S.Kim Wyner-Ziv Residual Video Codec WZ frames W X WZ Decoder WZ Encoder Xer • • Xer X’ Y Residual of a frame with respect to an encoder reference frame (Xer) is fed into a Wyner-Ziv encoder. To avoid drift, Xer should be replicable at the decoder. Since the decoder takes into account motion, Y is expected to be a better estimate of frame X than Xer. The Wyner-Ziv decoder uses both Y and Xer to calculate the reconstruction X’. [Aaron, Zhang, Girod, Asilomar 2002] 23 J.S.Kim Pixel-Domain Wyner-Ziv Video Codec Interframe Decoder Intraframe Encoder Slepian-Wolf Codec WZ frames Scalar Quantizer W Turbo Encoder Turbo Decoder Buffer Reconstruction W’ Request bits Side Y informati on Interpolation/ Extrapolation Key frames I Conventional Intraframe decoding Conventional Intraframe coding I’ [Aaron, Zhang, Girod, Asilomar 2002] 24 J.S.Kim Pixel-Domain Wyner-Ziv Video Codec After Wyner-Ziv Decoding Decoder side information generated by motioncompensated interpolation PSNR 24.8 dB 16-level quantization – 2.0 bpp 0 pixels in error PSNR 36.5 dB 25 [Aaron, Zhang, Girod, Asilomar 2002] J.S.Kim DCT-Domain Wyner-Ziv Video Codec Intraframe Encoder Interframe Decoder WZ frames Dk’ Dk W DCT Scalar Quantizer Turbo Encoder Request bits For each transform band k Buffer Turbo Decoder Recon W’ IDCT Side information Yk DCT Y Interpolation/ Extrapolation Key frames I Conventional Intraframe coding Conventional Intraframe decoding I’ [Aaron, Zhang, Girod, Asilomar 2003] 26 J.S.Kim Rate-Distortion Performance - Salesman Encoder Runtime Pentium 1.73 GHz machine B frame ~ 100% Interframe 100% 3 dB WZ DCT ~ 7% WZ Pixel ~ 6% 6 dB I frame ~ 18% Every 8th frame is a key frame Salesman QCIF sequence at 10fps 100 frames [Aaron, Zhang, Girod, Asilomar 2003] 27 J.S.Kim Salesman at 10 fps DCT-based Intracoding 149 kbps PSNRY=30.0 dB Wyner-Ziv DCT codec 152 kbps PSNRY=35.6 dB GOP=8 [Aaron, Zhang, Girod, Asilomar 2003] Conclusion Increase efficiency of DVC Reduce H(X) : simple ME/MC? Increase H(Y) : better interpolation/extrapolation Stronger correlation between X and Y. X encoder R decoder X Y X? Y? P(X/Y) 29 J.S.Kim Conclusion Distributed coding is a fundamentally new paradigm for video co mpression Slepian-Wolf encoding, is fundamentally harder for practical appli cations due to the general statistics of the correlation channel The rate-distortion performance of Wyner-Ziv coding does not yet reach the performance of conventional interframe coder It is unlikely that distributed video coding algorithm will ever beat conventional video coding schemes in R-D performance Many authors believe that distributed coding techniques will soon complement conventional video coding to provide the best overal l system performance and enable novel applications 30 J.S.Kim Research Plan (with M.S. Vidhya Murthy) 31 J.S.Kim Research Plan Plan done ■ Plan and achievements plan (Aug.2008 – June 2009) 8 9 10 11 12 1 2 3 4 5 6 ▣ Simulation Environment and survey WZ CODEC - survey WZ CODEC - joint simulation model of H.264 CODEC ▣ Survey of Channel Coding Algorithm - channel coding/decoding algorithm - analysis of channel source code ▣ Implementation of Wyner-Ziv CODEC - Performance comparison of Pixel- and Transform-domain - Investigate the statistical characteristics and distributions of the residual signal - Efficient encoding algorithm (Quantization/Entropy coding or Slepian-Wolf encoding/decoding) - Efficient generation of Side Information - Implementation of channel encoding/decoding algorithm - Rate control (or buffer control) algorithm Now 32 J.S.Kim Q&A Thank you 33 J.S.Kim