A performance evaluation of texture measures for image classification and segmentation using the cascade-correlation architecture

The performance of several texture-based measures is compared with respect to their ability to classify and segment images. Texture measures considered are: co-occurrence matrices, features derived from the Fourier spectrum and Gabor filters. The performance of raw pixel gray level values and gray level averages as classification features is also investigated. The cascade-correlation neural network architecture is used as a classifier. It was found that certain measures derived from the Fourier spectrum outperformed other types. The size of the fragments used for classification played a dominant role with respect to performance