Abstract

The inputs to high-level mechanisms (such as those responsible for boundary extraction, grouping, surface interpolation, and object identification) ultimately arise from the parallel and automatic processes at the front end of the visual system. Hence, any rigorous psychophysical analysis of high-level mechanisms requires either controlling for, or computing the effects of, the initial low-level mechanisms. In recent years, understanding of the front end has increased substantially, raising hopes for a more rigorous study of high-level mechanisms. To illustrate how models of the front end might be applied to the study of higher-level mechanisms, two examples will be considered: texture segmentation and visual search for texture regions. It will be shown that the front end can account for a rather large percentage of the variance in the performance of these tasks, suggesting that the high-level mechanisms are able to extract, with nearly uniform efficiency, the relevant information provided by the parallel front end. Thus, many aspects of texture processing might be adequately accounted for by combining current models of the front end with relatively simple models of the high-level mechanisms.

© 1992 Optical Society of America