New methods to improve the accuracy of the pel-recursive Wiener-based motion estimation algorithm

Accurate and robust motion estimation algorithms are essential in video coding, and also in many other applications. One method to estimate a dense motion field between two frames of an image sequence is the pel-recursive Wiener-based motion estimation algorithm. Its greatest drawback is the slow convergence at the image and object boundaries. Our paper focuses on this problem, and gives different solutions to speed up the convergence of the uniresolution and multiresolution pel-recursive motion estimation algorithm and to calculate an accurate motion field based on two or more partially inaccurate ones. It also reveals a side-effect, that can be utilized directly in motion-based segmentation of image sequences, which is also an important part of the object-based video coding algorithms