TEXTURE SEGREGATION, SURFACE REPRESENTATION AND FIGURE-GROUND SEPARATION

A widespread view is that most texture segregation can be accounted fo r by differences in the spatial frequency content of texture regions. Evidence from both psychophysical and physiological studies indicate, however, that beyond these early filtering stages, there are stages of 3-D boundary segmentation and surface representation that are used to segregate textures. Chromatic segregation of element-arrangement patt erns-as studied by Beck and colleagues-cannot be completely explained by the filtering mechanisms previously employed to account for achroma tic segregation. An element arrangement pattern is composed of two typ es of elements that are arranged differently in different image region s (e.g. vertically on top and diagonally on the bottom). FACADE theory mechanisms that have previously been used to explain data about 3-D v ision and figure-ground separation are here used to simulate chromatic texture segregation data, including data with equiluminant elements o n dark or light homogeneous backgrounds, or backgrounds composed of ve rtical and horizontal dark or light stripes, or horizontal notched str ipes. These data include the fact that segregation of patterns compose d of red and blue squares decreases with increasing luminance of the i nterspaces. Asymmetric segregation properties under 3-D viewing condit ions with the equiluminant elements close or far are also simulated. T wo key model properties are a spatial impenetrability property that in hibits boundary grouping across regions with non-collinear texture ele ments and a boundary-surface consistency property that uses feedback b etween boundary and surface representations to eliminate spurious boun dary groupings and separate figures from their backgrounds. (C) 1998 E lsevier Science Ltd. All rights reserved.