Illumination-Invariant Texture Classification Based on Self-Similarity and Gabor Wavelet

The appearance of a surface texture is strongly dependent on the illumination direction. This is why current state-of-art surface texture classification methods require multiple training images captured under a variety of illumination conditions for each class. This paper presents an inexpensive method for illumination-invariant texture classification based on self-similarity and wavelet transform. First, we train images for per class, which are captured under a variety of illumination conditions, to produce a similarity map based on self-similarity to represent this class. In allusion to each image in the database, We also employ a self-similarity map to represent the right image. For similarity map of the test images, which are most different from the training images (different illumination slants of the same texture), are transform by wavelet decomposition to extract texture feature and perform one-aganinst-one SVM algorithm for classification. We use a wide range of textures in the Pho-Tex database for the experiments to evaluate the performance of the proposed method. Although simple, the scheme has produced promising results.