Efficient motion compensation using multiple warped reference pictures based on feature tracking and segmentation

In this paper we present an efficient motion compensation method using warped reference pictures reflecting dynamic local warping motions in the video sequences. To extract various local warping motions between two scenes, we integrate the Kanade-Lucas-Tomasi (KLT) feature tracking algorithm and feature segmentation scheme. After extracting warping motion parameters, we generate multiple warped reference pictures containing each warping motion, which are inserted to the original reference picture list as an additional reference pictures. Experimental results shows that bit rate reduction can be achieved up to 20% in some sequence having zooming motion and average 6% bit rate reduction is yielded through the whole test sequences.