Research On Fast Motion Estimation For Video Coding

In recent years, video coding technology has made considerable development. Nowadays, H.264/AVC has been widely used in various aspects of the society. It has been showed that motion estimation is the core technology in video coding. As motion estimation occupies more than a half of the encoding time and has a great impact on the quality of video coding. It is very important for motion estimation to speed up the coding process and develop coding quality. Therefore, the motion estimation has been one of the focuses in video coding. In order to achieve high coding performance, new algorithms have been developed as follows.The conventional motion estimation based on genetic algorithm has good global optimization ability, but its high complexity enhances the cost of computation and storage, which increases the coding time. The traditional genetic algorithms usually use a very small number of the genetic iterations, which may reduce the search precision. In order to resolve the defects of long search time and low accuracy, this paper proposed an algorithm based on genetic algorithm and pattern matching. According to statistical property and prediction of the motion vector, termination strategies are designed in this algorithm. And the matching algorithm is also used to optimize the genetic search. Experiments dedicate that this algorithm maintains good quality in video coding, and also it greatly reduces the search points and coding time.To reduce the great computational burden in multiple reference frames motion estimation (MRF-ME), in this paper, a novel fast approach is proposed in which particle filter (PF) is introduced into MRF-ME. Inspired by the basic idea that the search strategy of PF in target tracking is similar with that in MRF-ME, the proposed approach is accomplished by three steps:1) crucial frames detection; 2) exhaustive search in the crucial frames; and 3) PF search execution in the multiple reference frames. Experimental results show that the proposed algorithm can reduce the computational complexity significantly meanwhile maintaining the coding efficiency almost identical to full search.