Complexity-Controllable Motion Estimation for Real-Time Video Encoder

Motion estimation (ME) is the main bottleneck and by far the most time-consuming module in real-time video coding application. Based on sufficient statistical analysis, a novel complexity-controllable ME algorithm is presented in this paper. The proposed algorithm consists of three effective stages: 1) initial stage with zero-motion detection, 2) predictive stage with early termination scheme, and 3) local refined stage by small diamond search. Furthermore, with given complexity constraints, a complexity-controllable scheme is introduced into the search process, which can achieve the preferable tradeoff between motion accuracy and ME complexity. Experimental results demonstrate that our proposed algorithm achieves similar performance with significantly reduced number of search points in comparison with some well-known ME algorithms, such as diamond search, hexagon-based search and enhanced hexagon-based search, etc. Moreover, due to its complexity-controllable feature, our algorithm can be adapted to various devices with a wide range of computational capability for real-time video encoder.