Multiresolution motion compensation coding for video compression

Motion estimation using block matching algorithm (BMA) is frequently used to reduced the temporal redundancy found in video coding. The performance of the mean squared error (MSE) is usually better than the mean absolute difference (MAD). However, the computational complexity is higher due to the squaring operations. In this paper, we propose a multiple candidates approach to shorten the gap between the two methods with slight increase in computational complexity. Application of the proposed algorithm to subband motion-estimation is studied and simulation results demonstrate that great improvement in performance can be obtained with similar computational complexity as the MAD measure.