A dynamic search range algorithm for variable block size motion estimation in H.264/AVC
Zhenxing Chen, Yang Song, Takeshi Ikenaga, Satoshi Goto · 2007
Motion vector prediction (MVP) candidates are frequently and widely used to predict the start position for motion estimation (ME). They are rarely utilized for other purposes such as calculating dynamic search range (DSR). This paper proposes a novel DSR decision method for variable block-size motion estimation (VBSME) in H.264/AVC by utilizing the information included by two MVP candidates. Experimental result shows that the proposed method can reduce the integer motion estimation (IME) time to less than 10 % of that variable block-size full search (VBS -FS) costs. Even compared with the recently proposed DSR method (Yu and He, 2002), the proposed method can also achieve an up-to- 50 % reduction of computation. Moreover, from the viewpoint of fast ME algorithms, it is also experimentally proved that the proposed method can be combined with any popular fast ME algorithm, such as UMHexagonS (Chen et al., 2006), to further reduce its computation.