Object Segmentation Based on Multiple Features for Low Bit Rate Video Coding

摘要

Video object segmentation for object based video coding according to MPEG-4 should be able to combine may features of image sequence to achieve high quality object segmentation. The significant features include color, motion vector and change information. This paper presents the object segmentation algorithm which image features are combined to use in segmentation process. A fast recursive shortest spanning tree (FRSST) algorithm is used to segment image to be different intensity regions because of a good performance in image segmentation and a fast algorithm. Motion vectors are estimated from consecutive frames and searched by thresholding hierarchical block matching (HBM) which want quite low computation and give acceptable results. After that many features will be integrated in segmentation decision process. Then, the postprocessing refines the final segmentation output. The results from some test sequences especially in video conference sequence have good performance and show interested-object boundary clearly.