Homomorphic Normalization-Based Descriptors for Texture Classification

Illumination variation is an essential trait in texture analysis, since the same texture can be surrounded with different illuminations, which can greatly affect the classification rate. This paper introduces a new approach to extract the texture features applying illumination normalization-based texture descriptors. The prime objective is to enhance the classification accuracy by normalizing the illumination of colour textures. Normalization of illumination is achieved by applying homomorphic filter. For feature extraction, two relevant approaches grey-level co-occurrence matrix (GLCM) and Laws' mask are utilized. Experiments are conducted for normalized co-occurrence and Laws' filter for colour images. Classification rates of the traditional GLCM and Laws' mask descriptors are included for baseline comparison. The effectiveness of the introduced techniques is assessed on three benchmark texture datasets, i.e. STex, VisTex, and ALOT. A k-nearest neighbour (k-NN) classifier is utilized to perform texture classification. Results show that the proposed approach has achieved higher classification rates and outperformed existing methods.