Switched prediction with dynamic decision rule

An adaptive predictor is proposed in which the decision rule takes into account the dynamic behaviour of a set of fixed predictors in an assigned neighborhood of the actual pel. The differences between the proposed rule and the most usual criteria are outlined. The results of the measurements show that the dynamic rule performances are at least on a level with the performances of the other criteria examined, and the point of view of predictor sensitivity to changes in image statistics, the dynamic rule may even be preferable.