WynerZiv Coding for Video: Applications to Compression and Error Resilience

1
Wyner-Ziv Coding for Video Applications to
Compression and Error Resilience
2
Overview

• Distributed Source Coding
• Intraframe Encoding with Interframe Decoding
• Systematic Lossy Forward Error Protection

3
Distributed Source Coding
Wyner-Ziv Theorem
Encoder
Decoder
4
Practical Distributed Source Coding

• Practical Codes
• Coset encoding Pradhan and Ramchandran, 1999
• Trellis codes Wang and Orchard, 2001
• Turbo codes
• Garcia-Frias and Zhao, 2001
• Bajcsy and Mitran, 2001
• Aaron and Girod, 2002
• LDPC codes Liveris, Xiong, and Georghiades,
  2002
• Applications
• Image and Video
• Pradhan and Ramchandran, 2001
• Liveris, Xiong, and Georghiades, 2002
• Jagmohan, Sehgal, and Ahuja, 2002
• Puri and Ramchandran, 2002
• Aaron, Zhang and Girod, 2002
• Sensor Networks Chou, Petrovic and Ramchandran,
  2002

5
Wyner-Ziv Video Codec
Wyner-Ziv Decoder
Wyner-Ziv Encoder
Reconstruction
X
Scalar Quantizer
X
Y
6
Wyner-Ziv Coding for Compression

• Compression for mobile video cameras
• Simple encoder
• Possibly complex decoder

Interframe Decoder
Intraframe Encoder
Wyner-Ziv Coding
Side Information
7
Intraframe Encoder - Interframe Decoder

• Limits reconstruction distortion based on
  quantizer coarseness
• Very simple encoder
• Turbo code can perform joint source-channel
  decoding
• Decoder controls rate and generates side
  information

8
Rate-PSNR Plots compared to H.263

• Foreman QCIF sequence
• Uniform 2, 4, 16 level quantizers
• Slepian-Wolf codec
• Rate 4/5 Turbo code
• Pelt10-3 25 pixels per frame
• Interpolation MC with symmetric motion vectors

9
Rate-PSNR Plots compared to H.263

• Carphone QCIF sequence
• Uniform 2, 4, 16 level quantizers
• Slepian-Wolf codec
• Rate 4/5 Turbo code
• Pelt10-3 25 pixels per frame
• Interpolation MC with symmetric motion vectors

10
Foreman sequence
After Wyner-Ziv Coding
Side information
16-level quantization (1 bpp)
11
Sample Frame (Foreman)
After Wyner-Ziv Coding
Side information
16-level quantization (1 bpp)
12
Carphone Sequence
Wyner-Ziv Codec 384 kbps
H263 Intraframe Coding 410 kbps
13
Wyner-Ziv Coding for Error Resilience

• Conventional Forward Error Correction (FEC)
• Protects the bit stream representing the video
  signal
• Lossless correction
• For graceful degradation, needs layered
  representation of video
• Systematic Lossy Forward Error Protection

14
Systematic Lossy Forward Error Protection
MPEG Encoder
MPEG Decoder with Error Concealment
S
S
Error-Prone channel

• Protects the original video waveform
• Lossy protection
• For graceful degradation, does not require
  layered representation of video

15
Results
Carphone CIF Sequence H.26L encoding at 1 Mbps 1
macroblock loss Error-free Wyner-Ziv bits 4 and
16 level quantization Rate 4/5 turbo code Pelt10-3
100 pixels per frame
16
Carphone Sequence
No Error Protection 1 macroblock loss 33 dB
With forward error protection of 1.5 bpp 1
macroblock loss 38 dB
17
Embedded Wyner-Ziv Codec
MPEG Encoder
MPEG Decoder with Error Concealment
S
S
Error-Prone channel

• Graceful degradation
• Does not require layered representation

18
Conclusions

• Wyner-Ziv coding for two video applications
• Intraframe encoder-Interframe decoder
• Very simple encoder
• Performs up to 2 - 7 dB better than H.263
  intraframe coding
• Systematic Lossy Forward Error Protection
• Protects the video waveform
• Backward compatible
• Can achieve graceful degradation without layered
  representation