Median structures for image sequence prediction

Principles of constructing predictors are discussed and two approaches to constructing multilevel structures are presented. Simulation results on coding gain, prediction signal entropy and transmission error propagation are presented and compared to an adaptive linear predictor. The results show that median structures can be used to build predictors that give comparable performance with linear ones. Robustness to transmission errors is gained by using median methods. Also, the design methods are intuitively easy and simple.<
>