Distributed Video Coding

Bernd Girod, Anne Margot Aagon and Shantanu Rane, Proceedings of IEEE, Jan, 2005

Presented by Peter

Outline

Introduction Foundations of distributed Coding Low-complexity video encoding Robust video transmission Conclusion

Introduction

Standards like MPEG and H.26x, the encoder exploits the statistic of the source signal

Efficient compression can also be achiebed by exploiting sources statistic – partially or wholly, at the decoder ONLY

It is the consequence of information-theoretic bounds established in 1970s By Slepian and Wolf for distributed lossless coding By Wyner and Ziv for lossy coding with decoder side

information The traditional balance of complex encoder and simple

decoder is essentially reversed

Foundations of Distributed CodingSlepian-Wolf theorem for lossless distributed coding Distributed compression refers to the coding of 2(or more)

dependent random sequence Each encoder sends a sends a separate bit stream to a single

decoder Decoder operates jointly on all incoming bit streams and thus

exploit the statistical dependencies

Foundations of Distributed CodingSlepian-Wolf theorem for lossless distributed coding Consider 2 statistically dependent i.i.d. finite-alphabet random

sequences X and Y Can do better with joint decoding (but separate encoding) Slepian-Wolf theorem establishes the rate region RX + RY ≥ H(X,Y), RX ≥ H(X|Y), RY ≥ H(Y|X) Surprisingly, the sum of rates RX + RY can achieve the joint

entropy H(X,Y), despite separate encoders for X and Y

Compression with decoder side information A special case of the distributed coding problem Side information Y is available at the decoder but

not at the encoder RY = H(Y) is achievable for encoding Y RX ≥ H(X|Y) , regardless of the encoder’s access to

side information Y

Rate-Distortion Theory for Lossy Compression with Receiver Side Information In 1970s, Wyner and Ziv extended Slepian and Wolf’s work

for lossy compression They showed that in the case of Gaussian

memoryless sources and mean-squared error distortion In 2003, S. Pradhan et al. showed that source

sequences X that are the sum of arbitrarily distributed side information Y and independent Gaussian noise

In 1996, Zamir proved that the rate loss is less than 0.5b/sample for general statistics and a mean-squared error distortion measure

0|| DRDR YXWZ

YX

0|| DRDR YXWZ

YX

Low-complexity Video Encoding Current video compression standard require much more

computation for the encoder than for the decoder (5-10 times) Well suited for broadcasting or for streaming VOD systems

Some applications require low-complexity encoders, e.g. wireless video sensors for surveillance, wireless PC cameras, mobile camera phones… etc.

The Wyner-Ziv theory suggests that individual frames can be encoded independently but decoded conditionally

Key frames are intra coded using conventional methods Non-key frames are intra coded using Wyner-Ziv encoder and

decode using Wyner-Ziv decoded with key frames as “side information”

Low-complexity Video Encoding Even if the receiver is another complexity-

constrained device, Wyner-Ziv can be used in conjunction with a transcoding architecture

Pixel-Domain Encoding

The simplest system that the authors have investigated Combination of a pixel-domain intraframe encoder and interframe

decoder system The decoder assumes the difference between the side

information and the original pixel are Laplacian distributed “Request-and decode” process is repeated used until an

acceptable probability of symbol error is researched Neither motion estimation and prediction, nor DCT and IDCT are

required at the encoder Requires 2 feedback shift registers and an interleaver Experiments on PIII 1.2Ghz machine

Average encoding runtime about 2.1ms/frame for the Wyner-Ziv scheme

36/ms/frame for H.263+ I-frame coding 227.0ms/frame for H.263+ B-frame coding

Pixel-Domain Encoding

Pixel-Domain Encoding

Transform-Domain Encoding

The authors theoretically studied the transformation of both the source vector and the side information

Block-wise DCT (4x4) is used and DCT coefficients are grouped into subbands

Similar to pixel domain, Laplacian residual model is assumed

Laplacian parameters are trained from difference sequences

A gain of up to 2dB over pixel-based system is observed

Transform-Domain Encoding

Pixel-Domain and Transform-Domain Encoding

Joint Decoding and Motion Estimation Joint decoding and motion estimation at the decoder A robust hash code word is sent to aid the decoder

in estimating the motion When motion exists, the block’s hash code is sent

along with the Wyner-Ziv bits Decoder performs motion search to generate the

best side information block from the previous frame 5-20% of the hash codewords are sent Substantially outperform conventional intraframe

DCT coding, still a gap relative to H.263+ interframe coding

Joint Decoding and Motion Estimation

Robust Video Transmission

Wyner-Ziv coding can be thought of as a technique which generates parity information to correct the “errors’ of the correlation channel

A source signal is transmitted over an analog channel without channel

An encoded version is sent over a digital channel as enhancement information

Reed-Solomon codes are used, only the parity symbols are transmitted to the receiver when error occurs

The authors refer the system as systematic lossy error protection (SLEP)

Robust Video Transmission

Robust Video Transmission

Robust Video Transmission

Robust Video Transmission

Conclusions

Distributed coding is a fundamentally new paradigm for video compression

Slepian-Wolf encoding, is fundamentally harder for practical applications 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

Its inherent robustness is a further attractive property, graceful degradation with deteriorating channel conditions can be achieved without a layered signal representation

It is unlikely that distributed video coding algorithm will ever beat conventional video coding schemes in R-D performance]

The authors believe that distributed coding techniques will soon complement conventional video coding to provide the best overall system performance and enable novel applications

Comments?