Modelling image redundancy

The authors use the information theory concept of rate of information transmission to define different models of image redundancy. First they apply information theory to a simple model considering an image as a set of isolated pixels. Then, in order not to neglect a pixel's spatial context, they consider the grey level vectors of cliques. They introduce a distance in the grey level space to solve the problem of insufficient statistics for small images. Their third model takes into account the neighbourhood of a pixel by introducing a Markov random field. They have to determine some parameters of the MRF in order to obtain sufficient statistics from standard satellite images, and they propose a measure based on a generalized Ising model. Finally, results for the proposed definitions are presented for some synthetic and a large variety of SPOT XS1, XS2 and XS3 image triples.