A spectral histogram model for textons and texture discrimination

Based on a local spatial/frequency representation, the spectral histogram of an image is defined as the marginal distribution of responses from a bank of filters. We propose the spectral histogram as a quantitative definition for textons. The spectral histogram model avoids rectification and spatial pooling, two commonly assumed stages in texture discrimination models. By matching spectral histograms, an arbitrary image can be transformed via statistical sampling to an image with similar textons to the observed. Texture synthesis is employed, to verify the adequacy of the model. Building on the texton definition, we use the χ²-statistic to measure the difference between two spectral histograms, which leads to a texture discrimination model. The performance of the model well matches psychophysical results on a systematic set of texture discrimination data. A quantitative comparison with the Malik-Perona model is given, and the biological plausibility of the model is discussed