Image Structure Models of Texture and Contour Visiblity
Geisler, W.S. and Perry, J.S

The perceptual mechanisms underlying texture and contour grouping/segregation play a dominant role in determining the visibility of targets in complex backgrounds. In most quantitative models of texture segregation the image is initially processed by channels selective along certain fundamental stimulus dimensions such as spatial frequency and orientation. These channels generally contain a nonlinearity, such as full-wave rectification, so that they signal the local contrast energy within the bandpass of the channel. Another stage of linear filtering, followed by a simple edge finding or thresholding mechanism, is then applied to the channel outputs to find the texture boundaries or regions. Although these channel-energy models have been successful in predicting texture segregation and discrimination performance for some classes of stimuli, there are large classes of stimuli that are readily segregated by human observers but which cannot be segregated by channel energy. The evidence suggests that more sophisticated models incorporating perceptual organization mechanisms will be required to predict human texture and contour segregation performance. This paper describes new experimental evidence, and a working model which, in principle, can account for a wider range of human segregation and grouping capabilities. The premise of the model is that the visual system typically extracts rich descriptions of local image structure, and that it uses these descriptions for subsequent segregation and grouping. The model contains physiologically-based low level mechanisms for extracting primitives, matching mechanisms for detecting structural similarity, and grouping mechanisms for binding structural parts into wholes. Quantitative predictions of the model for contour segregation performance are presented.