Purchase this article with an account.
Ross Goutcher, Pascal Mamassian; Selective biasing of correspondence matching in ambiguous stereograms. Journal of Vision 2003;3(9):317. doi: 10.1167/3.9.317.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Purpose: Constraints for resolving the stereo correspondence problem have been central to the success of models of stereovision. However, many such models use only one or two of the many matching constraints suggested in the literature. More specifically, models often utilise a nearest disparity (ND) constraint at the expense of a nearest neighbour (NN) constraint. Here we quantify the role of NN and ND constraints by putting them into competition.
Methods: We present a new wallpaper-type stereogram offering alternate percepts of a single, opaque surface or two-surface stereo transparency. By alternating the contrast of horizontally adjacent image dots we are able to selectively bias matching toward either of these percepts, which correspond to ND and NN matching, respectively. This allows us to quantify the relative strength of each constraint. Based on previous findings in the literature, we hypothesise that ND matching will exceed NN matching, and that NN matching will increase when such matches lead to the minimum disparity between surfaces.
Results: Points of subjective equality (PSE) between NN and ND matches were obtained for each observer. PSEs are biased towards the ND for most observers, indicating that the effects of ND matching exceed those of the NN. As hypothesised, the proportion of NN matches increases as the disparity between transparent surfaces decreases [cf. Zhang et al (2001), Vis Res, 41, 2995-3007].
Discussion: Utilising our new wallpaper stimulus we have selectively biased the correspondence matching process and obtained quantitative measures of the strength of matching constraints. This stimulus offers new possibilities for research into both correspondence matching and the bases of perceptual ambiguity. Our current results indicate that multiple constraints impinge on the stereo matching process and that ND matching is the strongest of these. We discuss how constraints may interact, and how our findings affect models of stereo matching.
This PDF is available to Subscribers Only