Hierarchical motion estimation using texture analysis

Motion estimation and compensation is used in image coding to reduce the prediction error for a larger reduction in the bit rate. In the past, various motion estimation techniques have been successfully applied to interframe predictive coding which can be broadly classified as pel recursive and block matching algorithms. In the block matching method the current frame of an image sequence is divided into blocks of equal sizes. Then for each block the motion vector is derived by minimizing a distortion measure, such as the mean square error (MSE), for all the possible search positions within a predefined search area (search window) in the previous frame. The hierarchical block matching algorithm used in this work is based on the image pyramid. The original picture frames, which are the lowest level in the resolution pyramid, consist of 320×256 pels each. The next level of the pyramid can then be formed by low-pass filtering the original image and subsampling the resultant picture by two in each dimension. In this presentation a 3 hierarchical level pyramid is used