Color Texture Classification under Varying Illumination

Color texture descriptors have gained a lot of interest in computer vision applications. Methods for grayscale texture analysis have been extended to color images. This paper presents a new descriptor for color texture analysis based on the Local Mapped Pattern (LMP) methodology called Color Local Mapped Pattern (CLMP). For each color channel, C-LMP considers the sum of the differences of each pixel of a given neighborhood to the central pixel as a local pattern that can be mapped to a histogram bin by using a mapping function. The histograms obtained from the color channels are concatenated in a color texture descriptor. The classification performance of the C-LMP is performed over Outex 14 texture database, considering three illumination sources. In our experiments, three different color spaces were considered: RGB, L∗a∗b∗ and HSV. Our results show that the C-LMP has better performance under illumination variances than the best results reported.