Abstract
Lateral inhibition is thought to underlie most forms of contextually induced repulsion. Examples include simultaneous contrast and the tilt illusion. It may seem strange that evolution has favoured a sensory apparatus that distorts the relationships between its stimuli, but it is possible that such distortion is accompanied by an increase in sensitivity. If this hypothesis is correct, then observers should be relatively insensitive to stimuli defined by feature dimensions in which repulsion does not occur. We searched for a dimension like this by systematically manipulating texture statistics. Luminance (a first-order statistic) obviously does not qualify, nor does contrast (a second-order statistic), but contrast-contrast (a fourth-order statistic) might. Orientation does not qualify, but orientation bandwidth might. Testing for repulsion between high-order texture values is difficult because the boundary between textures having different values is often indistinct. Might repulsion be predicted from boundary visibility? Not very well. Our re-analysis of Bosten and Mollon's (2010) data suggests a moderate correlation (r = 0.60) across 9 feature dimensions for population median indices of repulsion and sensitivity. On the other hand, our measurements, which confirm large repulsions across contrast-defined and orientation-defined texture boundaries that were twice their detection thresholds, revealed only inconsistent repulsion across equally visible boundaries defined by orientation-bandwidth. Using a d' of 1.4 to define the just-noticeable difference, there are only 4 or 5 noticeably different orientation bandwidths. We considered the possibility that there might be a correlation between this JND gamut and various indices of repulsion, but back-of-the-envelope calculations did not support this notion. Two further predicted consequences of lateral inhibition similarly failed to manifest in our laboratory: 1) a concomitant decrease of repulsion and discriminability with the distance between spatially separated textures (also absent from Mollon & Danilova, 2003) and 2) resistance from low-frequency masking (also absent from Westrick et al, 2013).
Meeting abstract presented at VSS 2014