New image segmentation method by modified counter-propagation network and genetic algorithm

We present a method for image segmentation using GA-based feature selection and neural net classifiers. We use a GA to select the optimal feature indices as the input of the neural net classifiers. Our GA method is based on an evaluation function, namely vector quantized conditional class entropy. By this measurement, we can evaluate the combination of feature indices rapidly without testing the actual classifiers. We use two types of neural net classifiers: the backpropagation network and a modified counter-propagation network. We applied our method to some classification problems and showed the effectiveness of our method.