Presentation is loading. Please wait.

Presentation is loading. Please wait.

Efficient Motion Vector Recovery Algorithm for H.264 Based on a Polynomial Model Jinghong Zheng and Lap-Pui Chau IEEE TRANSACTIONS ON MULTIMEDIA, June.

Similar presentations


Presentation on theme: "Efficient Motion Vector Recovery Algorithm for H.264 Based on a Polynomial Model Jinghong Zheng and Lap-Pui Chau IEEE TRANSACTIONS ON MULTIMEDIA, June."— Presentation transcript:

1 Efficient Motion Vector Recovery Algorithm for H.264 Based on a Polynomial Model Jinghong Zheng and Lap-Pui Chau IEEE TRANSACTIONS ON MULTIMEDIA, June 2005

2 Outline Error Concealment Motion Vectors Using Polynomial Model Motion Vector Recovery Using Polynomial Model Experimental Results Conclusions

3 Error Concealment (1) Recover lost information with spatially and temporally adjacent samples at the decoder  No additional bit rate  More computational complexity at the decoder Spatial information Temporal information ii+1 i-1

4 Error Concealment (2) Problem: Neighboring macroblocks are usually in the same packet Solution: Flexible macroblock ordering (FMO) in H.264/AVC Slice 0 Slice 1 Slice 2 Slice 0 Slice 1 Slice 2 Slice 0Slice 1 Used for ROI Used for concealment

5 Motion Vectors Using Polynomial Model Mv 1 Mv 2 Mv 3 Mv 4 Vl1Vl1 Vl2Vl2 Vl3Vl3 Vl4Vl4 Vr1Vr1 Vr2Vr2 Vr3Vr3 Vr4Vr4 Motion vector (y i )Vl1Vl1 Vl2Vl2 Vl3Vl3 Vl4Vl4 Mv 1 ? Mv 2 ? MV coordinate (x i )-22-18-14-10-6-2 0 -6-10-14-18-222610141822 e.g. lost

6 Motion Vector Recovery Using Polynomial Model (1) Minimize F(a 0, …, a m ) for j = 0, 1, …, m    e.g. Let m = 2 Motion vector (y i )Vl1Vl1 Vl2Vl2 Vl3Vl3 Vl4Vl4 Mv 1 Mv 2 MV coordinate (x i )-22-18-14-10-6-2  ‚ ƒ  ‚ ƒ

7 Motion Vector Recovery Using Polynomial Model (2) Recover the damaged block depending on slice types  Interleaving loss Reconstruct MV = (Mv v + MV h ) / 2 Reconstruct MV = Mv v  No interleaving loss Slice 0 Slice 1 Slice 0 Slice 1 Vertical approximation Horizontal approximation Vertical approximation Slice 2

8 Experimental Results Video sequence: Tempete, Foreman, Mobile, Stefan Frame format: QCIF GOP structure: IPPPIPPP… Comparisons: boundary matching algorithm  Choose the MV with the slightest blocking effect Mv 1 Mv 2 Mv 3 Mv 4

9 Experimental Results No error concealment Temporal replacement Boundary matching algorithm proposed With interleaving macroblock Tempete 5% 10%

10 Experimental Results Without interleaving macroblock Tempete 5% 10%

11 Conclusions Use an approximate function to describe the change tendency of motion vectors Can it be used for motion vector predictors?


Download ppt "Efficient Motion Vector Recovery Algorithm for H.264 Based on a Polynomial Model Jinghong Zheng and Lap-Pui Chau IEEE TRANSACTIONS ON MULTIMEDIA, June."

Similar presentations


Ads by Google