Image redundancy and classification

Abstract

C.E. Shannon in his Information Theory defined the rate of information transmission of a transmitter-receiver couple, also called system mutual information. This concept can be applied to a simple model considering an image as a set of isolated pixels in order to define an image redundancy measure. For applications where we want a redundancy measure which takes into account a pixel's neighbourhood, we propose two different approaches, one considering couples of neighbour pixels, and another based on the theory of Markov Random Fields (MRF). Finally we propose a method for unsupervised image classification based on the maximization of the redundancy of the classification and the image to classify.