Abstract
In brief displays, relative disparity facilitates the recognition of objects on natural backgrounds (Caziot, Valsecchi, Gegenfurtner & Backus, ECVP 2011). Is this because stereo attracts attention or because stereo specifies the object contour? To address this question we decoupled stereoscopic contours from luminance contours. Target objects had either object-defined disparity contours (ODDC) or were embedded in a rectangular local background (RDC). Performance should be better for ODDC if stereoscopic contours participate in the segmentation process. Stimuli were displayed on a rear-projection screen at 2m and fixation distance, defined by vergence demand, was 15m (approximately the distance at which background pictures were taken). Targets were 10ϊ2; wide, centered at one of 16 locations at 12ϊ2; eccentricity. Targets had 0, 15, 30, 60 or 120 arcmin of crossed disparity relative to fixation and the background, which portrayed a building façade. Displays lasted 33, 67, 100 or 133ms, followed by a mask. Subjects reported which of 6 possible targets was displayed. Performance increased with display duration and peaked at a disparity of 60 arcmin for both ODDC and RDC. The effect of disparity for RDC implies that disparity attracted attention, which facilitated recognition. However performance was significantly higher for ODDC than for RDC for all non-zero disparity values, but only at the longer display durations, which implies that disparity-defined object contours are useful for recognition, probably due to their contribution to segmentation. Thus, disparity facilitates image segmentation across a large range of disparities (2 deg or more) and even at short display durations (33.3 ms). Much of this facilitation results from attentional orienting, but disparity-defined contours also contribute, presumably to segmentation. Perhaps surprisingly stereoscopic contours can facilitate target recognition for diplopic targets.
Meeting abstract presented at VSS 2013