Rotation and illumination invariant texture analysis: Matching clothes with complex patterns for blind people

Matching clothes is a challenging task for blind people. In this paper, we describe a new texture analysis approach which is robust to variations of rotation and illumination for matching two images of clothes with complex patterns. The approach integrates Radon transform, wavelet transform, and co-occurrence matrix for texture analysis of clothes images. We first calculate directional properties of texture pattern in each image by applying Radon transform. The directional properties are used to rotate the clothes image with the dominant orientation of the texture patterns as horizontal. To handle illumination changes, we perform Haar wavelet transform to extract texture features on three directions (horizontal, vertical and diagonal). For each wavelet sub image, grey level cooccurrence matrix for texture analysis is calculated. Finally, texture matching is performed based on six statistical features (i.e. mean, variance, smoothness, energy, homogeneity, and entropy). The robustness and effectiveness of the proposed method are evaluated on our database which contains 128 images of complex pattern clothes. The matching results are presented for blind users as speech outputs.