Multi-resolution Laws' Masks based texture classification

Wavelet transforms are widely used for texture feature extraction. For dyadic transform, frequency splitting is coarse and the orientation selection is even poorer. Laws' mask is a traditional technique for extraction of texture feature whose main approach is towards filtering of images with five types of masks, namely level, edge, spot, ripple, and wave. With each combination of these masks, it gives discriminative information. A new approach for texture classification based on the combination of dyadic wavelet transform with different wavelet basis functions and Laws' masks named as Multi-resolution Laws' Masks (MRLM) is proposed in this paper to further improve the performance of Laws' mask descriptor. A k-Nearest Neighbor (k-NN) classifier is employed to classify each texture into appropriate class. Two challenging databases Brodatz and VisTex are used for the evaluation of the proposed method. Extensive experiments show that the Multi-resolution Laws' Masks can achieve better classification accuracy than existing dyadic wavelet transform and Laws' masks methods.