An adaptive LS-Based motion prediction algorithm for video coding

초록

In this paper, we introduce an adaptive motion vector prediction algorithm to improve the performance of a video encoder. The block-based motion vector can be characterized by the local statistics so that the coefficients of LS-based linear motion predictor can be optimized. However, it requires very expensive computational cost: which is major bottleneck in real-time implementation. In order to resolve the problem, we propose the LS-based motion prediction algorithm using spatially varying motion-directed property, so that the coefficients of the motion predictor can be adaptively controlled, resulting in the reduction of computational cost as well as the prediction error. Experimental results show the capability of the proposed algorithm.