1 / 21

Feature-Based Intra-/InterCoding Mode Selection for H.264/AVC

Feature-Based Intra-/InterCoding Mode Selection for H.264/AVC. C. Kim and C.-C. Jay Kuo CSVT, April 2007. Outline. Introduction Overview of Proposed Algorithm Feature Selection Feature Space Partitioning Coding Mode Prediction Experimental Results. Introduction.

baird
Download Presentation

Feature-Based Intra-/InterCoding Mode Selection for H.264/AVC

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. Feature-Based Intra-/InterCoding Mode Selection for H.264/AVC C. Kim and C.-C. Jay Kuo CSVT, April 2007

  2. Outline • Introduction • Overview of Proposed Algorithm • Feature Selection • Feature Space Partitioning • Coding Mode Prediction • Experimental Results

  3. Introduction • Inter/Intra Mode Decision in H.264 • Skip mode, direct mode, intra modes, and inter modes • Full mode decision • Testing all possible modes and then choosing the best mode with smallest cost • Fast algorithms • Selection of optimal inter-prediction mode • Selection of optimal intra-prediction mode • Binary decision of intra/inter mode

  4. Overview of Proposed Algorithm Motion activity Choose min(f0,f1) Risk-Free Compute risk- minimizing mode MB Risk-Tolerable Risk-Intolerable f0, Residual of inter prediction Full mode decision f1, Residual of intra prediction

  5. Feature Selection (1/4) • Intra mode feature • Calculate SATD for 5 modes • DC, vertical, horizontal, diagonal down-left, and diagonal down-right • Let f1 or fIntra be the SATD of the MB of the chosen modes

  6. Feature Selection (2/4) • Inter mode feature • MV is obtained by • MVFAST + Two more candidates • Residual of every visited point is remembered in the memory • Search points of a MB < 512 • Let f0 or fInter be SATD of MB residual of the chosen MV (i-1,j-1) (i,j) n-1 n

  7. Feature Selection (3/4) • Motion activity classification • Motion activity, decision error, and skipped frames • Decision metric • df = f1 – f0 • Intra (Inter): df < 0 (df > 0) • Decision error probability • P(df<0inter)+P(df>0intra)

  8. Feature Selection (4/4) • Motion activity, RD cost difference dc, and feature difference df • dc = (D1 +1R1) - (D0+ 0R0) • Positive (Negative) if inter (intra) is better Best intra mode Best inter mode Low motion medium motion High motion

  9. Feature Space Partitioning • The 3-D feature space is partitioned into three regions (off-line) • Lp: normalized RD cost between the best mode and the wrongly selected mode Threshold Motion activity Inter mode feature Intra mode feature

  10. Feature Space Partitioning • Let every cell has about equal training data |MV| f1 f0

  11. Feature Space Partitioning • Getting training data from Akiyo, Hall Monitor, Foreman, Coastguard, Stefan, Table Tennis, and Mobile.

  12. Coding Mode Prediction (1/4) • Risk-Free region • Distribution of f0 and f1 in a given motion class based on feature difference Risk free

  13. Coding Mode Prediction (2/4) • Risk-tolerable/-intolerable region Risk-tolerable and Risk-intolerable

  14. Coding Mode Prediction (3/4) Cost of deciding ~mi under mj • Risk-tolerable region • Risk function The chosen mode mj is the best mode m0: intra m1: inter For simplicity, let  stands for cost instead of R

  15. Coding Mode Prediction (4/4) • Risk-minimizing mode selection • Mode selection rule 0 0 • Likelihood ratio • Parametric • Semi-parametric • Nonparametric

  16. Experimental Results (1/6) • Environments • JM7.3a • 32 x 32 motion search range • Fast full search with 5 reference frames • No B-frame • QP= {10, 16, 22, 28, 34} • 5 skipped frames

  17. Experimental Results (2/6) • QCIF Table Tennis

  18. Experimental Results (3/6) • QCIF Table Tennis Saving time Computation complexity

  19. Experimental Results (4/6) • QCIF Foreman • 5 skipped frames 0 skipped frames

  20. Experimental Results (5/6) • QCIF Stefan

  21. Experimental Results (6/6)

More Related