A fast embedded selection approach for color texture classification using degraded LBP

We propose a fast embedded selection approach for color texture classification using Local Binary Pattern (LBP). This texture descriptor transforms an image by thresholding the neighborhood of each pixel and coding the result as a binary number. The selection approach presented in this paper is based on a degraded definition of the color LBPs. To compute these degraded LBPs, we take care of choosing a relevant reduced neighborhood - or a combination of reduced neighborhoods - with respect to the analysed textures. This leads to consider histograms with a lower dimension and so to reduce the computation times. We thus propose to determine the dimension of the selected feature subspace with these degraded color LBPs and to use this dimension for the classification with the classic LBPs. Experimental results carried out with benchmark databases in different color spaces show that this approach allows to obtain such good classification results than when the basic definition of LBP is used, while significantly reducing the learning time.