Nonlinear image processing for optimum composite source coding

Abstract

The concept of composite source coding for efficient image transmission or storage is addressed in the paper, and specific properties characterising the component sources are determined. An approach is developed which decomposes typical image signals into two independent additive components. Each component is subsequently encoded separately for storage or transmission. These individual components are combined at the receiver (or recovery stage) to produce a replica of the original signal. Special attention is focused on the optimality of the signal partitioning in the sense that the rate required to transmit the two components is minimum for a priori specified values of the average distortion in the reconstruction of the original signal. A formulation of the mathematical problem is effected in such a way that nonlinear techniques based on dynamic programming are used to achieve efficiently the optimum partitioning. Of the two components, only one is allowed a certain but controllable degree of degradation for the purpose of transmission rate reduction. The other component, which is associated with certain perceptually essential image features, is transmitted without distortion. The subjective quality of the reconstructed image is thereby preserved without incurring further increase in the total transmission rate.