distributed video coding bernd girod, anne margot aaron, shantanu rane, and david rebollo-monedero...

14
Distributed Video Coding Bernd Girod, Anne Marg ot Aaron, Shantanu Ran e, and David Rebollo-M onedero IEEE Proceedings 2005

Post on 22-Dec-2015

223 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Distributed Video Coding

Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero

IEEE Proceedings 2005

Page 2: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Outline

Foundations of Distributed Coding Low-Complexity Video Encoding

Page 3: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Foundations of Distributed Coding (1) What is distributed coding?

Coding of multiple dependent random sequences with separate encoders sending separate bit-streams to a single decoder.

Based Slepian-Wolf and Wyner-Ziv information-theoretic results

0 2 4

1 3 5

0 1 2 3 4

Encoder

Encoder

Decoder

Page 4: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Foundations of Distributed Coding (2) Slepian-Wolf Coder and Wyner-Ziv Coder

X and Y are very similar

Lossless coder

Side information

Page 5: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Foundations of Distributed Coding (3) Given two dependent i.i.d. random sequences X and Y.RX ≥ H(X), RY ≥ H(Y)

Slepian-Wolf theoremRX + RY ≥ H(X, Y)

RX ≥ H(X|Y), RY ≥ H(Y|X)

entropy

Joint entropy

encoding decoding

X

Y

X

Y

Page 6: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Foundations of Distributed Coding (4) Slepian-Wolf coding

Slepian-Wolf coder Encoder: Encoding X without Y Decoder: Reconstructing X with Y

Assumptions X and Y are very similar Y is known at the decoder

XChannelcoding XP

Y

PYP

Channeldecoding X

Parity bits

Alternative 2:encoder decoder

Slepian-Wolf encoder

Slepian-Wolf decoder

ABCABCABC

XA

ABCABCABC

YY

X

encoder decoder

Alternative 1:

Page 7: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Foundations of Distributed Coding (5) RD Theory for Lossy Compression with Re

ceiver Side InformationDistortionWyner-Ziv RD function

in the case of Gaussian memoryless sources and mean-squared error

distortion, or X is the sum of arbitrarily distributed Y and independent

Gaussian noise.

Y is known at the encoderY isn’t known at the encoder

Page 8: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Foundations of Distributed Coding (6) Wyner-Ziv Coding

Reconstruct with side information Y.Assumptions

Quantization step size Three interleaved quantizers: A, B, and C

A AB B CC

Y X

X

★ log23 bits

A AB B CC

X X A

Y

Encoder Decoder

(3/2)δ(3/2)δ

Page 9: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Low-Complexity Video Encoding (1) Conventional video encoder

5-10 times more complex than the decoder Suitable for the case that video is compressed once

and decoded many times Broadcasting or VOD systems

Distributed video encoder Low-complexity encoder, but high-complexity decoder Suitable for

Wireless video sensors for surveillance Wireless PC cameras Mobile camera phones Disposable video cameras

Page 10: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Low-Complexity Video Encoding (2) Pixel-Domain and Transform-Domain Encoding

A Laplacian distribution between S and is assumed

The Laplacian parameter is estimated from previous decoded frames

Encoding time (Pentimu III 1.2GHz) – pixel-domain encoding

Wyner-Ziv: 2.1 ms/frame H.263 I-frame: 36 ms/frame H.263 B-frame: 227 ms/frame

KeyFrame

W-ZFrame

W-ZFrame

W-ZFrame

KeyFrame

W-ZFrame

KeyFrame… …

S

Page 11: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Low-Complexity Video Encoding (3) Pixel-Domain and Transform-Domain

Encoding

Page 12: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Low-Complexity Video Encoding (4) Side information in decoder side

Copying from previous frames, motion-compensated interpolation, multiple frame predictors, …

e.g. Motions estimation at the decoder Additional information is helpful

CRC or some coefficients of the quantized symbol

CRC(0,0)

CRC(0,1)

CRC(0,2)

CRC(1,0)

CRC(1,1)

CRC(1,2)

CRC(2,0)

CRC(2,1)

CRC(2,2)

CRC(1,1)

previouscurrent

encoder decoder

Page 13: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Low-Complexity Video Encoding (4) Rate control

Controlled by the decoder Using a feedback channel Must be performed online useful information can help flexible generation of side

information through the feedback channel Controlled by the encoder

Classifying blocks into several modes with different rates Using the frame difference or block behavior

Better side information cannot lower the bit-rate Can be performed offline

Page 14: Distributed Video Coding Bernd Girod, Anne Margot Aaron, Shantanu Rane, and David Rebollo-Monedero IEEE Proceedings 2005

Low-Complexity Video Encoding (5) Some topics about DVC

How to generate side information?Spatial domain or frequency domain?What is the optimal quantizer for DVC?Rate control in DVCRobust transmission…