Abstract
Investigation of the spatial properties of filters in the human visual systems that process luminance contours has made extensive use of the optimal balance between spatial frequency content and spatial localization offered by 2D Gabor patches. We investigate whether these properties can be useful for probing the detection of contours defined in the motion domain. Motion defined contours are ecologically important cues to object boundaries in a three-dimensional world, and also can help to disambiguate camouflage. We designed motion-defined Gabor stimuli which consist of limited lifetime moving random dots. The velocity of the dots for any given position is defined by the 2D Gabor function, and dots can either be moving parallel or orthogonal to the motion defined contours. The Gabor patterns are embedded in motion noise dots with speeds drawn randomly from the speeds in the Gabor area. In a contour detection task participants were asked to judge whether the motion defined Gabor pattern is oriented horizontally or vertically. We find that observers become very proficient at performing this task, with as little as 12 dots providing enough information for 90% accuracy. We measured performance over different values of spatial extent of the Gabor function and found that the number of correct responses increases and saturates with size. We then chose the average point at 90% of peak performance, using a fitted function and used this fixed spatial extent to measure the number of correct responses at various spatial frequencies. We find that the number of correct responses decreases with higher frequencies, with optimal detection being at 0.1 cycles/degree, the lowest spatial frequency tested. These results suggest that putative human motion contour detectors may be most sensitive to single isolated edges. We also consider these results experimentally in the context of vernier acuity for motion contour localization mechanisms.
Supported by the Engineering and Physical Sciences Research Council