Gray level cooccurrence histograms via learning vector quantization

In this paper, we propose to use learning vector quantization for the efficient partitioning of a cooccurrence space. A simple codebook is trained to map the multidimensional cooccurrence space into a 1-dimensional cooccurrence histogram. In the classification phase a nonparametric log-likelihood statistic is employed for comparing sample and prototype histograms. The advantages of vector quantization are demostrated with a difficult texture classification problem involving 32 textures. We also point out two problems in the use of cooccurrence matrices that should be taken into account in order to achieve the best possible classification accuracy. Finally, we compare the performance of cooccurrence histograms to that of GMRF features and Gabor filtering, proving that gray level cooccurrences are a powerful approach if