DIRECT mode early decision optimization based on rate distortion cost property and inter-view correlation

Yun ZHANG, Sam KWONG, Long XU, Gangyi JIANG

Research output: Journal PublicationsJournal Article (refereed)peer-review

13 Citations (Scopus)

Abstract

In this paper, an Efficient DIRECT Mode Early Decision (EDMED) algorithm is proposed for low complexity multiview video coding. Two phases are included in the proposed EDMED: 1) early decision of DIRECT mode is made before doing time-consuming motion estimation/disparity estimation, where adaptive rate-distortion (RD) cost threshold, inter-view DIRECT mode correlation and coded block pattern are jointly utilized; and 2) false rejected DIRECT mode macroblocks of the first phase are then successfully terminated based on weighted RD cost comparison between 16×16 and DIRECT modes for further complexity reduction. Experimental results show that the proposed EDMED algorithm achieves 11.76% more complexity reduction than that achieved by the state-of-the-art SDMET for the temporal views. Also, it achieves a reduction of 50.98% to 81.13% (69.15% on average) in encoding time for inter-view, which is 29.31% and 15.03% more than the encoding time reduction achieved by the state-of-the-art schemes. Meanwhile, the average Peak Signal-to-Noise Ratio (PSNR) degrades 0.05 dB and average bit rate increases by -0.37%, which is negligible. © 1963-12012 IEEE.
Original languageEnglish
Pages (from-to)390-398
JournalIEEE Transactions on Broadcasting
Volume59
Issue number2
Early online date26 Apr 2013
DOIs
Publication statusPublished - Jun 2013
Externally publishedYes

Funding

This work was supported in part by the Hong Kong RGC General Research Fund (GRF) Projects 9041495 (CityU 115109) and the Natural Science Foundation of China under Grants 61102088, 61272289, and 61071120.

Keywords

  • Digital video broadcasting
  • DIRECT mode
  • early termination
  • mode decision
  • multiview video coding

Fingerprint

Dive into the research topics of 'DIRECT mode early decision optimization based on rate distortion cost property and inter-view correlation'. Together they form a unique fingerprint.

Cite this