Efficiency analysis and application of uncovered background prediction in a low bit rate image coder

A fast algorithm for gaining scene background information from a TV sequence is presented. This picture information is used by an uncovered background predictor in a motion-compensating predictive coder for videoconference or picture telephone applications. In uncovered background regions, prediction error entropy is reduced close to the value of the frame difference entropy in static regions, which is regarded as a practical lower bound. Almost all these picture elements can therefore be reconstructed without transmission. A motion-compensating hybrid differential pulse-code modulation (hybrid-DPCM)/transform coder is supplemented by uncovered background prediction. The maximum bit rate reduction achieved in the experiments is 9% for a picture where 14% of the picture elements of temporally changed regions belong to uncovered background regions