Texture Classification and Recognition Using a Composite Feature Set and Neural Network

摘要

A new composite texture feature set is proposed and used in texture classification and recognition with a three layers Backpropagation Neural Network. The experiments show that our texture feature set give excellent performance on a variety of textures. The recognition accuracy on verification set for the texture pairs and groups of four textures are 100% after the Neural Network being well trained. It was shown that the features do capture the essential texture characteristics, and be capable of classifying a wide variety of textures with high accuracy and computational efficiency. They are invariant under rotation, darkening or lightening. All these indicate that, this new feature set probably has a promising applicability for wide variety of image classification and segmentation applications.