Texture classification using a novel Walsh/Hadamard transform

This paper proposes a novel Walsh/Hadamard-based feature extraction approach to texture classification, called Directional Walsh/Hadamard Transform or DWHT. The DWHT algorithm, an implementation of multi-scale and multi-directional ideas in ordinary Walsh/Hadamard transforms, shows high efficiency in textural feature extraction. A comparative test applied on a database of VisTex textures, illustrates the advantages of the proposed method in both classification accuracy and the computational costs.