Statistical Characteristics of Simple Wyner-Ziv Frames Jin-soo KIM.

Statistical Characteristics of Statistical Characteristics of

Simple Wyner-Ziv FramesSimple Wyner-Ziv Frames

Jin-soo KIM

2 J.S.Kim

Contents

Simple Stat. Char.s of Wyner-Ziv Frames Statistical Characteristics of Simple Frame Difference DCT Coefficients for Simple Frame Difference Spatial Reduction for WZ Encoder Spatial Reduction and DCT for WZ Encoder

Research Plan Q&A

3 J.S.Kim

Statistical Characteristics of Simple Frame Difference

4 J.S.Kim

Simple Frame Difference for WZ frames

I I Iwz wz wz I

EncoderSide

- =Residualsignal

I Iwz

DecoderSide

Encoded bit stream

Recon.Residualsignal

MCFI SideInformation

WZDecoder

1 2 3 4

1 2 3

WZEncoder

[Aaron, Zhang, Girod, Asilomar 2002]

5 J.S.Kim

Frame Difference for Simple WZI I Iwz wz wz I

Residualsignal

6 J.S.Kim

Statistical Distribution

Luma

7 J.S.Kim

Standard Deviations for Simple FD

Akiyo: Luma's STD = 2.8, Chroma's STD = 0.6, Y/C entropy=1.69, 1.0bits/pel

Stefan: Luma's STD = 35.6, Chroma's STD = 7.4, Y/C entropy=6.2, 2.7bits/pel

Coastguard: Luma's STD = 16.4, Chroma's STD = 1.6, Y/C entropy=5.7, 1.6bits/pel

300 Frames @ CIF, 30Hz

Foreman: Luma's STD = 16.5, Chroma's STD = 3.4, Y/C entropy=5.2, 1.9bits/pel

8 J.S.Kim

DCT Coefficients for Simple Frame Difference

9 J.S.Kim

Frame Difference for Simple WZ

I I Iwz wz wz I

wz

8x8 DCT

Residualsignal

[Aaron, Zhang, Girod, Asilomar 2003]

10 J.S.Kim

Std Dev of DCT Coeff For Simple FD

Std. dev.

DCT coeff. in zig-zag scan order

11 J.S.Kim

Spatial Reduction for WZ Encoder

12 J.S.Kim

Blk Diagram for Spatial Reduction WZ Enc.

2:1 decimation2:1 decimation

Current frame(WZ frame)

2:1 decimation2:1 decimation

Rec. Frame store(Key frames)

Regular Frame

Coder (ex. H.264)

Regular Frame

Coder (ex. H.264) 2:1 interpolation2:1 interpolation

++

-

Residualsignal

LR bit stream

13 J.S.Kim

Block Diagram for Simulation

LPF & 2:1 decimation

LPF & 2:1 decimation 2:1 interploation & LPF 2:1 interploation & LPF

Integer QuantInteger Quant

-- ==Residualsignal

14 J.S.Kim

Std Dev for Residual Signal

Stefan: Luma's STD = 12.1, Chroma's STD = 1.0, Y/C entropy=4.7, 0.5bits/pel

Coastguard: Luma's STD = 10.4, Chroma's STD = 1.6, Y/C entropy=4.8, 1.0bits/pel

300 Frames @ CIF, 30Hz

Foreman: Luma's STD = 9.1, Chroma's STD = 1.0, Y/C entropy=3.8, 0.5bits/pel

Stefan: Luma's STD var = 35.6, Chroma's STD = 7.4, Y/C entropy=6.2, 2.7bits/pel

Coastguard: Luma's STD = 16.4, Chroma's STD = 1.6, Y/C entropy=5.7, 1.6bits/pel

Foreman: Luma's STD = 16.5, Chroma's STD = 3.4, Y/C entropy=5.2, 1.9bits/pel

15 J.S.Kim

Spatial Reduction and DCT for WZ Encoder

16 J.S.Kim

Spatial Reduction and DCT for WZ Enc.

2:1 decimation2:1 decimation

Current frame

2:1 decimation2:1 decimation

Rec. Frame store

Regular Frame

Coder (ex. H.264)

Regular Frame

Coder (ex. H.264) 2:1 interpolation2:1 interpolation

++

-

ResidualsignalDCTDCT

[HP report by Mukherjee, Machiavello, Queiroz, 2007]

LR bit stream

17 J.S.Kim

Std Dev of DCT Coeff For Spatial Reduction

Std. dev.

DCT coeff. in zig-zag scan order

18 J.S.Kim

Comparison of DCT and DCT+Spatial Reduction

Std. dev.

DCT coeff. in zig-zag scan order

19 J.S.Kim

Conclusion

Need to find an efficient statistical model for the residual signal

Need to look for entropy coding based on the trade-off between computational complexity and R-D performance

Need to implement the channel encoding/decoding algorithm

Need to develop an effective reconstruction scheme for side information

20 J.S.Kim

Research Plan (with M.S. Vidhya Murthy)

21 J.S.Kim

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

■ ■ PlanPlan and achievementsand achievements■ ■ PlanPlan and achievementsand achievements

Research Plan Plandone

Now

22 J.S.Kim

Q&A

Thank you