Research And Implementation Of Motion Estimation Based On Directional Filter Bank

Effective motion estimation can reduce the energy of residual frame and improve video compression performance. This thesis analyzes two different scheme of motion estimation: hierarchical motion estimation and motion estimation in transform- domain.Motion compensation-embedded zero block coding is video coding algorithm based wavelet. Because directional filter bank (DFB) has the same multi-resolution and perfect reconstruct nature with wavelet, this thesis investigates video coding by using 2-D quincunx filter bank with respect to two methods of motion estimation: (1) DFB used in hierarchical motion estimation; (2) apply DFB in motion estimation in transform-domain.In hierarchical motion estimation, DFB is invoked to get the spatial lowpass pyramids. At first coarse estimation is made at low resolution. At the next level, the search is carried over the higher resolution, and here the motion vector estimate from the top level is doubled for the initial position and the search window is shifted by that much amount. So this method ignores the odd motion vector. This thesis proposes a simple two dimensional surface method to solve this problem. The method we proposed improves the accuracy of motion estimation.We also apply 2-D quincunx filter banks in motion estimation in transform- domain. In transform-domain we combine the algorithm of motion estimation in wavelet-domain with the prediction method we proposed. This motion estimation in transform-domain results in better PSNR values in experiments.