Texture image segmentation by genetic algorithms

Recently, genetic algorithms (GAs) have been widely applied to various optimisation problems. One of their application areas is combinatorial optimization. In this paper, we formulate an image segmentation problem as a combinatorial optimization and propose a method to segment an image composed of various randomized textures by using GAs. After a considered image has been divided into many small rectangular windows with the same size, the 2D autoregressive model, fractal dimension, mean and variance extracted from the data in each small window are used as a feature vector of that window. After clustering the feature vectors, the clustering evaluation function is examined, based on the total variance of the feature vectors in each cluster. Then, in order to obtain the optimum value of the evaluation function, clustering optimization is performed recursively by using genetic operations, such as crossover, mutation and selection. In numerical examples, the efficacy of the proposed segmentation method is verified. Furthermore, we also explore hybrid algorithms of GAs and Kohonen self-organizing neural networks, and compare the processing time of the method using GAs and that using the hybrid algorithms