Abstract
A series of experiments were carried out examining factors influencing the ability to detect contour structure within noise. In each experiment the observers' task was to detect a target embedded in a field of 900 random gabors. The target was defined as one or more pairs of aligned gabors separated by a small distance. When the pairs are placed randomly in the image, but with the same orientation, detection is poor, requiring 20 pairs to be present for performance to be good. When the pairs are aligned to be consistent with a single straight contour, detection is very good even for just 3 or 4 pairs. This effect occurs despite the distance between the pairs being considerably larger than the average distance between the elements in the background and background gabors filling the gaps along the line. To examine potential cues to the grouping process underlying the enhancement in performance, the relationships between the elements within a pair and between the pairs in the image were manipulated. The effect of changing the similarity of orientation, contrast, and disparity of the gabors within each pair and between pairs was measured. Results suggest that detection is optimal only for targets that are in a straight line. Performance is relatively unaffected by contrast, although having the same contrast within a pair is better than opposite contrast. There is no effect of contrast difference between pairs. There is small evidence of disparity differences affecting performance.