An efficient block motion estimation method on CELL BE

In order to take advantage of the byte-type data parallelism in CELL BE's SIMD technique, this paper introduces a new algorithm, which can achieve much higher prediction accuracy and lower computational complexity compared with diamond search (DS) algorithm. Most of the conventional fast motion estimation algorithms just focus on reducing the number of search points. However, less search points is not equal to less computational complexity for CELL BE's architecture. The introduced method uses a large rectangular search pattern including 35 search points at the first search step. Through increasing search points, it gains 4.32% coding efficiency compared with DS. Through SAD reuse and reducing unaligned memory access, it needs 29.5% less computation than DS method on CELL BE. Our implementation of this algorithm on CELL BE shows that we can satisfy the requirement of real-time HD H.264 encoding.