抄録
This paper proposes a computational complexity reduction scheme for region-of-interest (ROI) based H.264/AVC encoding for videophone, video conferencing and surveillance systems. A proposed fast ROI detection algorithm which can detect face-like regions as ROI is applied to obtain accurate and small ROI so as to reduce the necessary coding effort of the encoder. The complexity reduction algorithm contains 3 methods: (1) the inter prediction mode selection based on quality difference, (2) unequal performance degradation based on unequal bits allocation, and (3) the ROI boundary enhancement to reduce the coding complexity of ROI. Experimental results show that proposed novel ROI detection and complexity reduction scheme can reduce 77.18% simulation time when QP difference between ROI and non-ROI is 20, with 1.02% bit-rate increase and 0.04dB PSNR decrease. 18.17% of encoding time is further reduced compared with previous work with similar performance degradation.
