Abstract: This paper proposes a novel and simple fast Block-Matching Algorithm (BMA), called Adaptive Variety-Pattern Search (AVPS), which efficiently utilizes the high relation of adjacently block’s motion vectors. The algorithm adopts a series of skills, which are composed of prediction of initial search point, effective half-stop criteria, adaptive search pattern and auxiliary search points. Extensive experiments conducted based on the H.264 encoding platform show that the proposed AVPS can reduce the computational complexity by 16%~33% and 30%~50% compared with the UMHexagonS and EPZS, while maintaining similar PSNR and bit rate.

Key words: motion estimation, prediction of initial search point, half-stop criteria, adaptive search pattern

摘要： 充分利用运动矢量的时空相关性，提出一种新的块匹配运动估计算法，称为自适应可变模式搜索算法。该算法充分利用搜索起点预测、中止准则、自适应搜索模板、辅助点搜索等一系列技术。H.264编码平台上的实验表明，该算法可以保持高信噪比和低比特率，其计算复杂度只有UMHexagonS的16%~33%和EPZS的30%~50%。

关键词: 运动估计, 搜索起点预测, 中止准则, 自适应搜索模板

CLC Number: 

• TP391