Image Segmentation Using Excess Entropy

Abstract

We present a novel information-theoretic approach for thresholding-based segmentation that uses the excess entropy to measure the structural information of a 2D or 3D image and to locate the optimal thresholds. This approach is based on the conjecture that the optimal thresholding corresponds to the segmentation with maximum structure, i.e., maximum excess entropy. The contributions of this paper are several fold. First, we introduce the excess entropy as a measure of the spatial structure of an image. Second, we present an adaptive thresholding method based on the maximization of excess entropy. Third, we propose the use of uniformly distributed random lines to overcome the main drawbacks of the excess entropy computation. To show the good performance of the proposed segmentation approach different experiments on synthetic and real brain models are carried out