Abstract
We investigated the role of contour integration in a task that required observers to identify an object at different positions in the visual field. Our goal was to measure the ability to match shapes presented sequentially in different parts of the visual field (translation invariance) as a function of shape-element grouping. Contours consisted of seven discrete 'Gabor' elements that were either aligned to form a smooth path (aligned with the direction of the path) or a jagged path (randomly oriented with respect to the path). Observers were presented with smooth or jagged contours either side of the fixation point (2 deg, 4 deg, or 6 deg, to the left or right) for 250 ms. Following an ISI of 250 ms, a second contour was presented, also for 250 ms. For one half of the trials, the same contour was presented in both intervals; for the other, a small change was made to the contour path. Also, on half of the trials, both contours were presented in the same position, and on half the trials the second contour was presented in the opposite hemifield, resulting in a separation distance of 4 deg, 8 deg, and 12 deg. A staircase procedure estimated threshold. Our results show that the process of contour integration can significantly improve observers' ability to match paths (i.e., determine if they are the same of or different) across the visual field. Three control experiments demonstrated that this improvement is due to binding of the contour-elements, and not due to a local orientation process. Furthermore, we found that observers were quite accurate at matching paths across relatively large distances with only a small fall-off in performance with eccentricity. The results are discussed in the context of underlying processes involved in translation invariance.
Meeting abstract presented at VSS 2012