Abstract
Surfaces in the world can produce extended contours on the retina. How are these extended contours perceptually constructed by visual processes that initially extract only small pieces of the contour (e.g., neurons with small receptive fields in V1)? We used response classification to determine how the orientations of small (12 arcmin) edge elements along an extended semi-circular contour influenced the perception of that contour. Four observers were familiarized with two canonical patterns formed by placing 13 elements either A) tangent to a semi-circular contour (3.2 deg diameter), or B) orthogonal to this contour. Observers perceived A as a smooth, semi-circular contour. Pattern B resembled a “necklace” with the elements pointing radially to the center of the semicircle. We then independently and randomly perturbed the orientations of the 13 elements using a uniform “orientation noise” distribution (±90 degrees in 5 degree bins) to create 2560 stimuli (each presented twice). Observers made a forced-choice classification on each trial; did the stimulus (presented for 167 ms) look more like the semi-circular or the radial canonical pattern? Using logistic regression we derived the relative weights for each observer for all 13 elements; the weights estimate the independent contribution of each element to the overall pattern classification. All observers gave more weight to elements near the middle of the stimulus (e.g., near clock positions from 5 to 7 for a 3 to 9 o'clock semi-circle). Using a model of contour integration based on natural image statistics (Geisler et al., 2001), we found that local elements were not treated completely independently in these judgments. Rather, adjacent elements were grouped into more extended pieces of the contour possibly using a transitive rule as proposed by Geisler et al. Even in a task that did not explicitly require contour integration, we find evidence for these elemental grouping processes.