TEXTURE CLASSIFICATION WITH SINGLE AND MULTIRESOLUTION COOCCURRENCE MAPS

We have developed methods for the classification of textures with mult idimensional co-occurrence histograms. Gray levels of several pixels w ith a given spatial arrangement are first compressed linearly and the resulting multidimensional vectors are quantized using the self-organi zing map. Histograms of quantized vectors are classified by matching t hem with precomputed texture model histograms. In the present study, a multiple resolution technique in linear compression of pixel values i s evaluated. The multiple resolution linear compression was made with a local wavelet transform. The vectors were quantized with the tree-st ructured variant of the self-organizing map. In the tree-structured se lf-organizing map, the quantization error is reduced, in comparison to the traditional tree-structured codebook, by limited lateral searches in topologically-ordered neighborhoods. The performance of multiresol ution texture histograms was compared with single-resolution histogram s. The histogram method was compared with three well-established metho ds: co-occurrence matrices, Gaussian Markov random fields, and multire solution Gabor energies. The results for a set of natural textures sho wed that the performance of single- and multiresolution texture histog rams was similar. Thus, the benefit of multiresolution analysis was ov erridden by the multidimensionality of our texture models. Our method gave significantly higher classification accuracies than the three oth er methods.