High Precision of Global Motion Estimation in Multi-dimensional Transform Domain

Abstract

Based on the multi-dimensional vector matrix discrete cosine transform (MVM-DCT), this paper presents a global and accurate motion estimation in multi-dimensional transform domain in order to solve the problem that the precision of the block matching motion estimation algorithm is discrete and the block matching is a local optimum .The algorithm breaks the traditional spatial and temporal limits, and adopts the dimension reduction and the linear iterative fitting with dynamic intelligent windows. The algorithm ensures the accuracy continuity and global optimality of the motion estimation. Experiment results verify the correctness of the theory. The accuracy can reach the magnitude of-4 10 and the speed of the convergence of linear iterative fitting is very fast.