Image Segmentation by Region Growing Using Cartesian Genetic Programming

Abstract

In this paper, we propose a method of image segmentation using Cartesian Genetic Programming. Image segmentation is a significant challenge, and many methods have been proposed. However, the results of segmentation are different from each purpose, and it is difficult to segment images into regions requested by different users in different ways respectively. In our method, we segment images by region growing method into regions as same as requested by user. Region growing segments images by merging neighboring regions whose similarity are high, however we could not find criteria of similarity in the phase of merging regions. Therefore, we use the outputs of the functions generated by Cartesian Genetic Programming, one of the structure optimization algorithm, and optimize the structure of them fitting for each purpose. We use these functions for calculating similarity between regions. Image features are used for inputs, and outputs are used for similarity between regions. We verified our method with images and their targets, and obtained the successful results of segmentation in several images.