Hierarchical motion estimation based on visual patterns for video coding

Abstract

Block matching algorithms(BMAs) are often employed for motion estimation(ME) in video coding. Most conventional fast BMAs treat the ME problem as an optimization problem and suffer heavily from the problem of being trapped at local minima. The full search algorithm(FS), on the other hand, is very time-consuming. Few of them makes use of the information inherent in the images explicitly. We propose a new ME algorithm which can reduce the search range while guaranteeing global optimality in most cases, making use of the edge features. Microblock visual patterns are designed to extract edge information to guide block matching: searching is only carried out at places where the real match most likely happens. The motion field subsampling technique is further employed to get a hierarchical algorithm, which can further double the speed. The proposed algorithms obtain speeds about ten times faster than that of FS with comparable prediction quality.