Exploiting scene structure in video coding

In video coding, information about scene structure can be estimated from previous frames in the video sequence. A technique has been proposed by the author (1990) which exploits such information about scene structure to improve motion compensation accuracy, without transmitting structural information to the decoder. The more accurate motion compensation has been shown to increase overall coding efficiency by 20-35%. The author reviews this approach to motion compensation and proposes a recursive (in time) algorithm for computing the required motion field segmentation. The algorithm reduces the computational complexity of motion field segmentation while increasing its reliability