Adaptively asymmetric pattern search for fast block-matching motion estimation

Abstract

In this paper, we invent a novel fast block-matching algorithm (BMA) for motion estimation (ME), called adaptively asymmetric pattern search (AAPS). Different to the other BMAs, this algorithm uses an asymmetrical search pattern and a predicted vector at the first search step, and dynamically adjusts the size of search pattern to minimize the number of search points. We have already applied AAPS to the process of motion estimation in the MPEG-4 encoding platform, and compared this algorithm with the other BMAs. The experimental result shows that AAPS achieves a 56-65% and 20% improvement in search speed respectively compared to the diamond search algorithm and the adaptive rood pattern search algorithm, and simultaneously maintains a high PSNR.