Image Segmentation Based on Maximum Entropy and Kernel Self-Organizing Map

This paper proposes a segmentation method based on information theory. The entropy of the image is regarded as the objective function to be optimized. It is maximized during the segmenting process. At first, the kernel self-organizing map is applied to cluster the input vectors of the image into groups according to their attributes, and it keeps entropy maximization meanwhile. Then the clustering result is partitioned using the maximum entropy principle. Finally, the image is segmented according to the partition of clusters. Objective evaluation methods are applied to assess the performance of the method. Experimental results show that this method has the advantages of fault tolerance and adaptability, and it can separate salient objects from the background correctly.