DCT-Based Motion Estimation

Abstract

A new motion estimation approach, the DCT-Based Motion Estimation Scheme (DXT-ME) utilizing the sinusoidal orthogonal principles to estimate displacements of moving objects in the transform domain, based upon the concept of pseudo phases, is presented in this paper. The computational complexity of this method is only O(N2) for an N x N block in comparison to the O(N4) complexity of Full Search Block Matching Approach (BMA-ME). In addition, the DXT-ME algorithm has solely highly parallel local operations and this property makes parallel implementation feasible. Furthermore, incorporation of DXT-ME with a video coder using DCT can combine the DCT and motion estimation algorithm to achieve further saving in overall system complexity and increase the system throughput. Unlike the pel-recursive algorithm, this scheme is robust for even very noisy images. Due to its feature matching property, we can employ simple preprocessing in images of complicated scenery to extract the features of moving objects for DXT-ME to further improve its performance. Finally simulation on a number of video sequences is presented to compare DXT-ME with BMA-ME.