Drastic differences in binocular disparity tuning of V4 cells for random dots and solid figures: Quantitative analysis and mechanisms

Abstract

A random-dot stereogram (RDS) contains many identical dots that give rise to numerous false matches for binocular correspondence. Thus, the determination of stereo matching is much more difficult when viewing an RDS than when viewing a solid-figure stereogram (SFS). Psychophysically, adaptation to left-right reversing spectacles differentially affects stereoscopic depth perception for figural stereograms than RDSs (Shimojo and Nakajima, 1981). In order to characterize the cortical processing required for stereoscopic depth perception when viewing RDSs and SFSs, we studied the responses of single neurons in area V4 to dynamic RDSs and SFSs (bars) in an alert, fixating monkey. Consistent with a previous study (Hegde and Van Essen, 2004), the binocular disparity tuning obtained with RDSs and SFSs differed markedly. Quantitative analysis revealed that the modulation amplitude, the tuning sharpness, and the preferred disparity of tuning curves tested with RDSs did not correlate with those of the SFSs, and the disparity tuning was sharper for RDSs than for SFSs. A possible cause of these differences may be the positional shift of monocular images that inevitably occurs with disparity changes in SFSs, but not in RDSs. We examined whether neuronal sensitivity for the positional shift of monocular images of SFSs could explain the differences in disparity tuning. In order to compensate for the response modulation caused by the monocular features of SFSs, we subtracted the linear sum of responses to the left and right monocular images from the binocular responses at each corresponding disparity. Despite this compensation, however, the differences between tuning curves obtained with the RDSs and SFSs remained essentially unaffected. A single read-out mechanism for V4 responses to the two types of stereograms would produce significantly different stereoscopic depth judgments. Our results suggest that stereo processing for RDSs and SFSs is not based on a single mechanism.