台大心理系

In stereopsis, the visual system combines a pair of horizontally shifted images from the two eyes to generate a 3D percept. The shift between the left and the right eye images is called binocular disparity. The perceived depth from disparity, which can be derived by the geometry of the scene alone, is generally considered to be independent from luminance contrast. Here, we show that luminance contrast does affect stereo perception. Such effect cannot be explained by the existing models of stereopsis for they cannot predict an effect of luminance contrast on depth perception. Instead, we proposed a multi-stage model that involves a contrast gain control followed by the disparity averaging to account for the luminance contrast effect on perceived depth. In this study, we conducted four psychophysics experiments to examine each part of the multi-stage model in order to understand the role of luminance contrast in disparity processing. First, we used random-dot stereograms as stimuli to test the effect of luminance contrast on perceived depth. Our results suggested a profound nonlinear luminance contrast effect on perceived depth from disparity. We then varied the depth modulation frequency to investigate the effect of 3D surface configuration on perceived depth. Our results showed that both the modulation frequency and the number of modulation cycle had an effect on perceived depth from disparity. We further manipulated the interocular luminance contrast difference in the random-dot stereogram to explore the binocular summation in depth perception. Our results demonstrated that the perceived depth was determined by both luminance contrast level and interocular luminance contrast difference. This suggested that there must be an early contrast gain control mechanism before the luminance contrasts in the two eyes add up. Finally, we changed the spatial frequency of the band-pass noise patterns to test whether the spatial frequency of the test images affect the perceived depth. Our results showed that (a) perceived depth in band-pass noise patterns also depend on luminance contrast, and (b) the limitation of the spatial frequency information in band-pass noise patterns did not restrict the perceived depth. This indicated that the disparity channels must have a broadband spatial frequency tuning. All these results can be accounted for by the contrast gain control mechanism and disparity averaging operation in the multi-stage model. Thus, we demonstrated the important role of luminance contrast in disparity processing and offered a neurophysiological plausible model that may benefit future studies.