Image segmentation using genetic method to select feature indices

This paper presents a new method to select the optimal feature indices based on a genetic algorithm (GA) for the problem of image segmentation. For improving the performance of image segmentation using neural-net classifier, it is important to select the effective combination of feature indices, among many candidates, to be used for the classification. In our genetic method, we use a new criterion, VQCCE, to evaluate the combination of feature indices rapidly without testing on the actual classifiers. We applied our method to problems of texture classification and brain MRI segmentation.