Combining region-based differential and matching algorithms to obtain accurate motion vectors for moving object in a video sequence

Motion estimation plays an important role in image processing, since temporal information has been regarded as a promising feature for both image segmentation and video coding. In this paper a hybrid approach is proposed to integrate a differential (gradient-based) optical flow approach and region-based matching approach to search for accurate object motion vectors. Our method adopts the Horn-Schunck optical flow constraint, in conjunction with several proposed techniques to convert the dense optical flow field to region-based motion field, and thereby suppress noise. The region-based matching approach is a modified version of the traditional block-matching algorithm, so that it can operate in region-based mode, and thereby enhance the visual effectiveness near the edges. Therefore, the proposed hybrid method has the tendency to obtain the estimation of "true" object motion inherited by the gradient-based approach, and also the superior visual effectiveness inherited by the block-matching approach.