Abstract
In this paper, motion compensation with segmentation is proposed. First, an image is uniformly divided into rectangular blocks as in BMA (Block Matching Algorithm) and a set of affine transform parameter estimated for each block is clustered by k-means algorithm. Next, each pixel is allocated to cluster based on prediction error culculated using affine transform parameters for each cluster. After eliminating small regions, segmented image is obtained. Computer simulation for the proposed object-based motion compensation prediction is performed, and simulation results are compared with that of BMA.
