Experiments in segmenting texton patterns using localized spatial filters

摘要

In this paper we apply a previously developed model of texture segmentation using multiple spatially and spectrally localized filters, known as Gabor filters, to the analysis of textures composed of elementary image features known as textons. It is found that for regularly- or irregularly-spaced texton patterns, the segmentation approach works well, in the sense that it is in accordance with visual segmentation. Differences in texton spacing, size, orientation, and phase are all found to lead to successful segmentations.