1 / 17

Fine-granular Motion Matching for Inter-view Motion Skip Mode in Multi-view Video Coding

Fine-granular Motion Matching for Inter-view Motion Skip Mode in Multi-view Video Coding. Haitao Yanh, Yilin Chang, Junyan Huo CSVT. Outline. Motivation Introduction of motion skip mode Methodology Experimental results Conclusion. Motivation.

marlin
Download Presentation

Fine-granular Motion Matching for Inter-view Motion Skip Mode in Multi-view Video Coding

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Fine-granular Motion Matching for Inter-view Motion Skip Mode in Multi-view Video Coding Haitao Yanh, Yilin Chang, Junyan Huo CSVT

  2. Outline • Motivation • Introduction of motion skip mode • Methodology • Experimental results • Conclusion

  3. Motivation • Global disparity can not well describe the inter-view corresponding relations in different image regions. • Fine granularity is introduced to obtain mare accurate motion information. Akko&Kayo, 640*480, 30fps

  4. Introduction – Motion Skip Mode • Use global disparity vector to search for the corresponding macroblock. • Motion information is derived from the corresponding MB in the picture of neighboring view.

  5. Introduction – Motion Skip Mode (cont’d) • Assume there is one inter-view reference picture and one temporal reference picture: • Iv,t : the picture in view v at time t • Bv,t : a 16×16 block in Iv,t • ,where V and Vref denote the coding view and the reference view • , where T and Tref denote the time instance of the coding picture and the reference picture

  6. Introduction – Motion Skip Mode (cont’d) • Use mean absolute difference(MAD) to evaluate the matching error: • h, w: height and width of coding picture • accuracy: 16-pel w w (x, y) (x, y) (x, y) h h reference frame coding frame

  7. Fine-granular motion matching • In H.264/AVC, 8×8 block is the basic unit to perform MC. • To estimate 8-pel accuracy global disparity vector between the coding picture and the inter-view reference picture. • DG: global disparity vector • XG, YG: x and y component of global disparity vector • S : search range with 8-pel accuracy • where

  8. Fine-granular motion matching (cont’d) • After the estimation of DG, we need to find the optimal disparity of the coding macroblock BV,T. • A search window of (4×8-pel) ×(4×8-pel) centers at (x+xG,y+yG) • Each × sign indicates a search point. 8*8 MB 16*16 MB

  9. Fine-granular motion matching (cont’d) • The 16×16 block centers at each search point, (x+xG+Δxi, y+yG+Δyi) for the ith search point. • Each 16×16 block is composed of four 8×8 blocks, {bi,j|j=1,2,3,4}. • Each 8×8 block bi,j has its own motion information mi,j, Mi ={mi,j |j=1,2,3,4} 8*8 MB 16*16 MB

  10. Fine-granular motion matching (cont’d) • Disparity vector at each search point is represented as: • To find the optimal disparity Dopt, Lagrangian cost function is employed: where • Mi is the motion information of the block BVref,T(i) at the ith search point. • DREC(Mi) is measured as the sum of the squared differences (SSD) between the original MB and the reconstructed MB. • RREC(ΔDi ) is the sum of bits to encode the whole MB and ΔDi.

  11. Fine-granular motion matching (cont’d) • To lower the complexity, the cost function, instead, is replaced for fast RD performance evaluation: where • Mi(x,y)x and Mi(x,y)y denote the MV components at x and y direction. • λMOTION = λMODE • The optimal motion information Mopt can be obtained once ΔDopt is determined.

  12. Fine-granular motion matching (cont’d) • In case there are multiple inter-view reference pictures, the optimal incremental disparity ΔDopt and the index kopt of the selected reference picture can be obtained with: where ΔDi,k represent the incremental disparity at the ith search point in the kth inter-view reference picture.

  13. Experimental environment • JMVM V5.0 • Test sequences: • QP: 22,27,32,37 • Search rage of disparity estimation: 96 • Size of the search window for the proposed fine-granular motion matching algorithm: (10×8-pel) × (10×8-pel)

  14. Experimental results • Ration of motion skipped MBs:

  15. Experimental results (cont’d)

  16. Rate-distortion comparison • With/without base view: Without base view With base view

  17. conclusion • 8-pel precision motion matching is applied to inter-view reference pictures. • Results show that the proposed algorithm increase the number of motion skip MBs. • Further improvement on overall RD performance for MVC can be achieved.

More Related