Variable block sized motion segmentation for video coding

In this paper, we present a new approach to variable block sized motion segmentation (VBSMS) in which the image is adaptively partitioned into blocks of variable size with homogeneous motion. Instead of the usual quadtree segmentation, a segmentation strategy is employed which provides higher scene adaptation with little transmission overhead. A new homogeneity criterion is proposed which is based entirely on motion information and not on gray-level. Results are presented for standard QCIF images which demonstrate that under the same SNR constraints, there is significant improvement of the compression ratio achieved by this approach over the conventional fullsearch fixed size and variable sized quadtree block matching methods.