The role of local features in shape discrimination of contour- and surface-defined radial frequency patterns at low contrast

Shape processing involves a progression from local to global analysis. A key aspect of this is the binding of distributed local features into an overall form followed by the extraction of the shape independently of its local contrast and spatial scales, so enabling the shape to be encoded based on its proportions without reference to its exact size or retinal location. Here we use contour- and surface-defined radial frequency (RF) patterns in a shape discrimination task, previously thought to reflect a global processing stage that has reached contrast and scale invariance. We compare performance across different spatial scales for a wide range of RF patterns (contour spatial frequencies of 0.7-10.0cpd, pattern radii of 0.5-10.5°), and sharp- and smooth-edged surface-RF patterns, all at low contrast (5× detection threshold). We show that shape discrimination thresholds for RF patterns have a complex series of dependencies on stimulus size (radius), contour spatial frequency (thickness) and contrast, with no scale invariance. Our results are at odds with earlier work showing no effect of radius and spatial frequency on discrimination thresholds. We show that this discrepancy can be accounted for by a differential effect of contrast on shape discrimination, with shape invariance only stabilizing at higher contrasts (10-20× threshold).