1. Rate-distortion Optimized Mode Selection Based on Multi-path Channel Simulation Markus Gärtner Davide Bertozzi Project Proposal Classroom Presentation 6th February 2001

2. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 2 Overview Hybrid Video Coding Mode selection Previous works Multi-path channel simulation: Architecture Distortion Measure Expected Results Workplan

3. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 3 Motion-compensated hybrid coder Intraframe DCT coder Motion compensated predictor Intraframe Decoder Mode Control XORXOR Decoder Encoder

4. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 4 Inter- / Intra-frame coding P Frame (inter): low bit rate, exploits temporal redundancy sensitivity to error propagation I Frame (intra): high bit rate, no temporal dependency stops error propagation I Frame P Frame

5. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 5 Optimal Mode Selection Salesman Foreman Packet error rate [%] Intra [%] 2 46 10 20 30 Source: Färber, Stuhlmüller, Girod; ICIP 1999

6. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 6 Previous Approaches Feedback based methods Transmission delay limits applicability Heuristic refresh frequency: periodic intra- coding of: Whole frames ( Turletti-Huitema ) Random blocks ( Coté-Kossentini ) Threshold methods (Liao-Villasenor, Färber-Steinbach- Girod) Content adaptive methods (Haskell-Messerschmitt) Rate-distortion optimization (Coté-Kossentini)

7. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 7 Block mode chosen according to Error propagation only beyond one frame Distortion measure as simple sum of D q and D c Block-weighted Distortion Estimate Coté-Kossentini

8. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 8 Distortion is calculated for each pixel Computational complexity Holds for inter-pel accuracy only Recursive optimal per-pixel Estimate SNR Miss America GrandmaSalesman Mother & Daughter CarphoneForeman ROPE37.8 dB35.4 dB33.6 dB32.8 dB29.9 dB26.7 dB BWDE37.2 dB34.2 dB31.6 dB30.7 dB28.1 dB25.0 dB Zhang-Reghunatan-Rose

9. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 9 Our Approach Coder Decoder Channel 1 Distortion Estimate Mode Selection Decoder Channel 2 Decoder Channel n

10. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 10 H.263 Coding Standard I-frame: DCT coding of each 8x8 block P-frame: DPCM, 8x8 DCT coding of error, one motion vector per macroblock Mode selection on macro-block basis frame 16x16 macroblock 8x8 block GOB

11. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 11 Channel Model Model on macro-block basis Channel 1 Channel 2 Channel n X X X X X Group of blocks “Controlled Randomness”

12. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 12 Distortion Measure d i comprises any distortion incurred by path i R D where Our approach:

13. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 13 Expected Results Channel modelling Cote-Kossentini H. 263 Average PSNR at decoder Error probability

14. Markus Gärtner, Davide Bertozzi: Robust Video coding Stanford University, 6th February 2001 14 Workplan Week 1Week 2Week 3Week 4Week 5 Literature Investigation Setup of H.263 Implementation of Channel models Performance measurements Final presentation