Adaptive Gabor filters for texture segmentation

This paper describes robust hierarchical modeling of the image amplitude spectrum via sets of bivariate Gaussian functions which involves: adaptive determination of a low-pass filter, clustering of residual high-pass spectrum, and parametric encoding of separate spectral segments. Based on this modeling a small set of Gabor filters tuned to the channel of high activity in the image Fourier spectrum is determined and used to generate feature images for texture segmentation. In the segmentation algorithm a similar robust modeling procedure is applied to encode histograms of the feature images as mixtures of univariate Gaussians.