Sub-Optimal Quarter-Pixel Inter-Prediction Algorithm (SQIA)
Hoi-Ming Wong, Oscar C. Au, Jinxin Huang, Shiju Zhang, Wei Yan · 2006
Motion estimation (ME) is an important part of modern video coding systems to exploit temporary redundancy in a video. Motion estimation is typically per-formed firstly with integer-pixel accuracy and then at sub-pixel accuracy, which includes half-pixel and quarter-pixel accuracy. When sophisticated fast integer-pixel accuracy motion-search algorithms are used to decrease the number of search points for integer-pixel motion search, quarter-pixel motion search becomes another important processing bottleneck in the encoding process. The conventional method is to search 8 half-pixel positions around the motion vector (MV) obtained from integer-pixel motion search, then do motion search in the same way on 8 quarter-pixel positions around the MV obtained from the half-pixel motion search, therefore, in total, 16 search points are needed. The proposed algorithm, named sub-optimal quarter-pixel inter-prediction algorithm (SQIA), successfully optimizes the quarter-pixel motion search part and improves the processing speed with low PSNR penalty.