Abstract
Glass patterns, constructed by overlaying a random dot pattern and a spatially transformed version, provide ideal stimuli for characterizing the visual system's ability to process form information at several levels. In detecting form, the visual system must first resolve the orientations of individual correlated dot-pairs and subsequently integrate relative orientations to yield perception of a global pattern. To investigate the visual system's ability to integrate and segregate form information based on pure color variations, we have manipulated the chromaticities of elements in Glass patterns. Such abilities are germane to image segmentation under illumination conditions in normal visual environments.
All experiments measured observers' ability to detect global patterns (concentric or translational) as a function of the percentage of properly oriented dot-pairs. For dot-patterns defined by a single chromatic excursion from isoluminant backgrounds: i)correlation thresholds are similar to those for luminance-defined dots; and ii)an equivalent-contrast metric approximately equates thresholds among chromaticities. For mixed-chromaticity experiments, perceptually equating contrast along differing color-axes was essential. To probe early-level form processing, we varied chromaticities of component dots within a dot-pair, observing: i)the ability of visual mechanisms to extract oriented dot-pairs decreased with increasing chromaticity differences; and ii)average bandwidths were similar for cardinal and intermediate directions in an opponent color space. Experiments, where paired-dots had the same color but chromaticity differed among dot-pairs, probed higher-level mechanisms. Here, even for large chromatic differences, the visual system: i)does not segregate noise dot-pairs from correlated dot-pairs on the basis of chromatic differences, and ii)appears to integrate oriented dot-pairs of differing chromaticities in forming a global percept. These results, true for both concentric and translational correlations, differed markedly from experiments which varied luminance contrast and polarity of the dot elements.
This work was accomplished in collaboration with J.A. Wilson and R.L. DeValois and was funded by NSF IBN-0116895 to ES.