Purchase this article with an account.
Isamu Motoyoshi, Fred Kingdom; Perception of global textures based on joint-orientation statistics. Journal of Vision 2009;9(8):1007. doi: https://doi.org/10.1167/9.8.1007.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Texture information is useful for segmenting objects and identifying different types of natural textured surfaces. A number of psychophysical studies suggest that rapid segmentation of abutting textures is primarily determined by differences in 2nd-order image features such as local orientation and size. Here, we examine the role of higher-order orientation statistics in the perception of a single texture region. Human observers were presented with textures comprised of adjacently paired Gabor elements with a particular orientation difference and relative angular position, but with absolute orientation randomized, and were required to discriminate these textures from ones consisting of random, i.e. spatially uncorrelated orientations. Observers were particularly sensitive to textures containing paired elements that formed collinear curves, parallel lines, V, T, and L shapes. These features are similar to Julesz's textons but are also characterized in terms of the degree of collinearity. We find that textures consisting of two populations of features such as curves and V shapes, but not straight lines and curves, often look like a random texture. This indicates opponent interactions among specific features, e.g. collinear (lines and curves) vs. non-collinear (parallels and V shapes) patterns. In support of this notion, we introduce a novel aftereffect in which a random texture appears to be structured following adaptation to a structured texture. For example, adaptation to a texture of randomly-oriented V shapes makes the subsequently viewed random texture appear to contain many smooth curves. We also demonstrate an analogous simultaneous-contrast illusion. These findings support the existence of visual mechanisms that process global texture-surface information on the basis of texton-like orientation statistics or collinearity.
This PDF is available to Subscribers Only