DocumentCode :
572930
Title :
A horizontal data reuse approach for fractional motion estimation in H.264/AVC encoder
Author :
Sun, Minmin ; Li, Li ; Wang, Jiawen ; Pan, Hongbing ; Li, Wei
Author_Institution :
Key Lab. of Adv. Photonic & Electron. Mater., Nanjing Univ., Nanjing, China
fYear :
2012
fDate :
24-26 Aug. 2012
Firstpage :
821
Lastpage :
825
Abstract :
Fractional Motion Estimation (FME) with quarter-pixel accuracy improves the video encoding efficiency significantly. The 6-tap 2-D FIR of half-pixel interpolation requires redundant access to search window SRAMs. Since the size of search window SRAMs is relatively large, the redundant access causes much extra power consumption. In previous works, the vertical data reuse was fully achieved, while the horizontal data reuse was not. In this brief, a horizontal data buffering approach is proposed to improve the horizontal data reuse. Through this approach, the reduction of access to the reference pixel SRAMs for FME of a macro-block is 45% for inter-modes with sub block size of 16×16 and 16×8, and 30% for inter-modes with sub block size of 8×16, 8×8 and 8×4. The reduction of dynamic power consumption caused by memory access is proportional to the reduction in memory access. The hardware overhead of this proposed approach is 5.62k gates, only 2.6% of the overall hardware cost.
Keywords :
SRAM chips; buffer storage; interpolation; motion estimation; power consumption; search problems; video coding; 6-tap 2D FIR; FME; H.264/AVC encoder; dynamic power consumption reduction; fractional motion estimation; half-pixel interpolation; hardware cost; hardware overhead; horizontal data buffering approach; horizontal data reuse approach; memory access; quarter-pixel accuracy; redundant access; reference pixel SRAM; search window SRAM; vertical data reuse; video encoding efficiency; Finite impulse response filter; Interpolation; data reuse; fractional motion estimation; redundant memory access; style;
fLanguage :
English
Publisher :
ieee
Conference_Titel :
Computer Science and Information Processing (CSIP), 2012 International Conference on
Conference_Location :
Xi´an, Shaanxi
Print_ISBN :
978-1-4673-1410-7
Type :
conf
DOI :
10.1109/CSIP.2012.6308979
Filename :
6308979
Link To Document :
بازگشت