Application of composite source models to lossless image compression

Abstract

It is difficult to model images using a single source model. We develop two composite source models for images. When these models are used for lossless image compression, the composite source models are shown to perform better than the traditional single source model in the sense of reducing the source modeling entropy. The main difference between these two models is the way the switching information is handled. In the nonadaptive composite source model, the switching information is transmitted as side information. While in the adaptive composite source model, the switching information is derived from the transmitted data, removing the need for side information.