Fast hierarchical search block matching algorithm by using multiple motion vector candidates

Abstract

In this thesis, we have proposed an improved multi-resolution motion search algorithm, namely, multi-resolution motion search algorithm using multiple candidates and spatial correlation of motion fields (MRMCS) that is effective in terms of both performance and VLSI implementation, and extend it so as to cover field-based ME as well as frame-based ME. The main concept of the MRMCS is to combine both spatial correlation of motion vectors and the merit of hierarchical search or multi-resolution search using multiple candidates in a computationally simple way, maintaining a similar rate-distortion performance to that of full search block matching algorithm. MRMCS can avoid the local minimum problem by using multiple candidates in the middle level; one among the causal motion vectors of spatially adjacent blocks and two predict predicted from the coarsest-resolution level based on the predefined distortion measure. The former and the latter contribute to finding a true motion vector in the region having continuous and discontinuous motion field, respectively. Furthermore, an adaptive prediction technique is of use to increase the accuracy of a candidate predicted from the causal motion vectors of spatially neighboring motion vectors. In computer simulations, it is proved that the proposed algorithm had some effective features appropriate for real-time motion-estimator, especially with large search area. That is, 1) MRMCS reduced the computation for finding motion vectors to 1.0% compared with full search block matching algorithm, and this complexity reduction ratio is close to minimum bound for a realizable motion estimator. 2) MRMCS achieved the motion estimation performance close to upper bound. Compared with the full search block matching algorithm, the maximum performance degradation reach to only -0.2 dB, which did not affect on the subjective quality of reconstructed images al all.