Adaptive object tracking and video compression with dynamic contours

Abstract

We present a novel video compression system resulting in very low bit rates, using dynamic object oriented compression. Our method separates important objects from the background and from trivial details, so as to keep good visual quality of the objects of interest. The resulting bit rate can be significantly reduced by sacrificing information on less relevant parts of the scene, or from portions of the video sequence which are of less interest to the end users. The method is based on an active dynamic contour algorithm which tracks the external contour of designated objects, and which we improve to provide more accurate dynamic tracking. The performance of the object tracking algorithm, as well as of our compression scheme are presented. Experimental results concerning bit rates and video quality are provided and compared with respect to the standard H.263.