Feature extraction of color texture using neural networks for region segmentation

The feature extraction of color texture by neural networks is studied. The purpose of this processing is to segment interesting regions from their background. However, there is a fundamental problem that the backpropagation neural networks have the function of a nonlinear discriminant analysis. We examine about this fact through some experiments. Two kinds of neural networks are used according to the previous research, and several results including the correct rate of discrimination have been obtained. Subsequently we perform similar experiments based on the linear discriminant analysis. Comparing these results, it is shown that the neural networks have the best performance and several convenient properties - the most interesting one is the ability to select an optimum shape of window for extracting texture features