Abstract
Segmenting the visual scene into distinct surfaces is one of the most basic aspects of visual perception. In natural scenes, adjacent surfaces often differ in mean luminance, which provides an important boundary segmentation cue. However, mean luminance differences between two surfaces may occur without any sharp change in albedo at their boundary, but instead arise from differences in the proportion of small light and dark texture elements within each surface. Here we investigate the performance of human observers segmenting such “luminance texture boundaries”. Luminance texture boundaries were synthesized by placing different proportions of white and black Gaussian micropatterns on opposite sides of a boundary whose orientation was left-oblique (-45 deg. w.r.t. vertical) or right-oblique (+45 deg.), and observers identified the boundary orientation in a 2AFC psychophysical task. We demonstrate that a model based on a simple luminance difference computation cannot explain observers' boundary segmentation performance. However, extending this one-stage model by adding contrast normalization successfully accounts for these data. By performing further experiments in which observers segment luminance texture boundaries while ignoring super-imposed luminance step boundaries, we demonstrate that the one-stage model, even with contrast normalization, cannot explain psychophysical performance. However a Filter-Rectify-Filter (FRF) model, positing two cascaded stages of filtering, fits this data very well, and furthermore can account for observers' ability to segment luminance texture boundary stimuli, both in the presence as well as absence of interfering (masking) luminance step boundaries. We propose that such multi-stage luminance difference computations may be useful for boundary segmentation in natural scenes, where shadows often give rise to luminance step edges which do not correspond to surface boundaries.