Abstract
Earlier work has demonstrated the benefits of grouping along a contour: patches in a roughly collinear string are easily detected amidst noise patches of random orientation. This study examines whether contour grouping interferes with the ability to detect changes in a roughly collinear string of patches.
The test patch was the end element of a 4 Gabor patch string and was either collinear with the string (0) or was tilted by 15, 30, or 45 with respect to the string. The string appeared at a random location within a 2 degree square aperture centered on fixation. In a 2IFC task, observers chose the interval with a contrast increment on the test patch. The contrast of the other elements in the string was 50%.
When the string was presented by itself, increment thresholds were highest for the collinear configuration and decreased with increasing tilt of the test patch from the string. However, when the same high contrast string was presented among randomly-oriented noise patches thresholds showed the opposite trend: thresholds were lowest when the patch was collinear with the string and increased with tilt. Collinearity also helped when the string was presented alone at low (10%) contrast. The benefits of collinearity were much reduced when the test patch appeared in a known location, suggesting a role for uncertainty as well as for collinear facilitation.
These results are consistent with single unit responses in striate cortex of cats and macaques (Polat et al, Nature 1998; Kapadia et al. PNAS 1999) showing that collinear interactions are suppressive when the stimulus is clearly visible (high contrast string without noise), whereas the collinear string both facilitates and acts as a cue to the test patch under conditions of poor visibility (noise or low contrast).