On the discrimination of micropatterns and textures

The Perceptron-type of approach of texture perception argues for decomposition of the patterns by detectors or filters which can interact to result in classification of spatial information. We consider a case of texture discrimination where textures composed of discriminable grey-level micropatterns may or may not be distinguished. Two types of models involving linear pattern decomposition are shown to be insufficient for explaining such results. This suggests that observers encode the signals by means of signal dependent or adaptive feature extractors. Such a concept of texture perception is related to connectionist ideas of information processing as pursued by Julesz in his texton theory.