H.264/AVC fractional motion estimation engine with computation reusing in HDTV1080P real-time encoding applications

Yang Song*, Ming Shao, Zhenyu Liu, Shen Li, Ngfeng Li, Takeshi Ikenaga, Satoshi Goto

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contribution

9 Citations (Scopus)

Abstract

H.264/AVC fractional motion estimation (FME) engine for HDTV 1080p is proposed in this paper. In order to provide real-time processing capability with reasonable hardware cost, several techniques have been presented. Firstly, the H.264/AVC is optimized and only 1 reference frame and block modes above 8 × 8 are supported. Therefore, the computation is reduced to 11.4% and the PSNR loss is only 0.1dB. Secondly, the lossless inside-mode and cross-mode reusing techniques are adopted, which can reduce about 65% pixel generation and SATD calculation. Thirdly, the lossless optimized FME scheduling is used to remove the pipeline bubbles between adjacent 1/2-pel and 1/4-pel FME. The proposed FME engine is realized with TSMC 0.18μm 1P6M CMOS technology and costs 203.2K gates and 52.8KB SRAM. Under 200MHz frequency, the proposed FME engine can real-time encode HDTV 1080p at 30fps with 236mW power cost.

Original languageEnglish
Title of host publication2007 IEEE Workshop on Signal Processing Systems, SiPS 2007, Proceedings
Pages509-514
Number of pages6
DOIs
Publication statusPublished - 2007 Dec 1
Event2007 IEEE Workshop on Signal Processing Systems, SiPS 2007 - Shanghai, China
Duration: 2007 Oct 172007 Oct 19

Publication series

NameIEEE Workshop on Signal Processing Systems, SiPS: Design and Implementation
ISSN (Print)1520-6130

Conference

Conference2007 IEEE Workshop on Signal Processing Systems, SiPS 2007
Country/TerritoryChina
CityShanghai
Period07/10/1707/10/19

ASJC Scopus subject areas

  • Media Technology
  • Signal Processing

Fingerprint

Dive into the research topics of 'H.264/AVC fractional motion estimation engine with computation reusing in HDTV1080P real-time encoding applications'. Together they form a unique fingerprint.

Cite this