Adaptive feature-based motion analysis and its application to interframe coding

Contemporary methods of interframe coding use motion estimation and compensation as a means of reducing the prediction error in algorithms which provide a large reduction in data rate. These methods can broadly be divided into two categories, pel recursive and block matching algorithms. Further work has investigated model-based coding, using the displacement and deformation of image objects as the criteria for interframe coding. Consideration of a typical audio-visual scene indicates that a motion search of the entire image is unnecessary, since motion is mainly concentrated in the detailed area around the subject´s face. The technique described can be considered as a step between the simple full-search block matching method and the more involved approach of model-based coding. It is demonstrated that, using a low-resolution image, objects can be segmented into features of consistent texture, small enough to provide interframe detail and large enough to provide the basis of an algorithm of low complexity and real-time processing potential