1. Wyner-Ziv Coding of Motion Video

2. Overview Intraframe encoding with Interframe decoding
Wyner-Ziv Video Codec
Simulation Results

3. Interframe Video Compression
Current video standards
Interframe predictive coding for compression
Encoder is 5-10 times more complex than decoder
X’i-1
Interframe Encoder
Interframe Decoder Xi
Xi’
Standard codec

4. Intraframe Encoding – Interframe Decoding
Dual System
Simpler encoder
Possibly complex decoder
Interframe Decoder
Intraframe Encoder Xi
Xi-1’
Xi’
Proposed codec
Wyner-Ziv Coding
Side Information

5. Slepian-Wolf and Wyner-Ziv 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]
Practical Applications
[Pradhan and Ramchandran, 2001]
[Jagmohan, Sehgal, and Ahuja, 2002]

6. Wyner-Ziv Video Codec X’ X Y Even frames are encoded independently
Intraframe Encoder
Interframe Decoder
Slepian-Wolf Codec
Even frame X
Scalar Quantizer
Turbo Encoder
Turbo Decoder
Reconstruction X’ Y
Buffer
Request bits
previous
Interpolation
Odd frames
next
Even frames are encoded independently
Odd frames are known as side information at the decoder

7. RCPT-Based Slepian-Wolf Codec
Even frame X
Decoded quantized
symbols
Scalar Quantizer
Turbo Encoder
Turbo Decoder
Buffer
Y2i
Request bits Y
Uniform scalar quantizer – no coset grouping
RCPT Slepian-Wolf Codec
Flexibility for varying statistics
Embedded puncturing pattern
Bit rate controlled by decoder through feedback

8. Side Information Y Interpolation
previous
Interpolation
Odd frames
next
Interpolation
Average same block from previous and next frame
Motion-compensated interpolation with symmetric motion vectors
Flexibility in design for decoder
Statistics between side information and current frame
Laplacian residual model
Estimate the Laplacian parameter at the decoder

9. Reconstruction Function
Decoded quantized
symbols
Reconstruction X’ Y
Limits the magnitude of the reconstruction distortion
Need dithering to avoid contouring
Pixels reconstructed independently

10. Simulation Quantizer 2, 4, 16 levels Slepian-Wolf codec
Rate 4/5 Turbo code
Embedded puncturing pattern with period 8
Pe<10-3 ~ 25 pixels per frame
Interpolation
Averaging
MC with symmetric motion vectors (SMV)
Rate-PSNR Comparison with H263+
Intraframe coding
Interframe coding (B frames) with no motion compensation
Interframe coding (B frames) with motion compensation

11. Carphone Sequence
6 dB
2 dB
8 dB

12. Foreman Sequence
7 dB
4 dB

13. After Wyner-Ziv Coding
Foreman sequence
Side information SMV Interpolation
After Wyner-Ziv Coding
16-level quantization (~1 bpp)

14. After Wyner-Ziv Coding
Sample Frame
Side information SMV Interpolation
After Wyner-Ziv Coding
16-level quantization (~1 bpp)

15. After Wyner-Ziv Coding
Sample Frame
Side information Average Interpolation
After Wyner-Ziv Coding
16-level quantization (~1 bpp)

16. Carphone sequence Wyner-Ziv Codec SMV Interpolation 384 kbps
H263+ Intraframe Coding 410 kbps
Wyner-Ziv Codec SMV Interpolation 384 kbps

17. Conclusion Use Wyner-Ziv coding for practical compression application
Used statistics of the source
New video system
Intraframe encoder – Interframe Decoder
Compared to H263+
2 to 7 dB better than Intraframe coding
5 to 8 dB worse than Interframe coding with MC
Further improvements
Exploit spatial correlation
Acceptable symbol error rate