Purchase this article with an account.
Arash Yazdanbakhsh, Oliver Layton, Ennio Mingolla; A Neural Model of Figure-Ground Segregation Explains Occlusion Without Junction Detectors. Journal of Vision 2011;11(11):1174. doi: https://doi.org/10.1167/11.11.1174.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
T-junctions, created by the intersection of contours separating three different luminance values, are often said to provide local information about occlusion in visual scenes. Although neurophysiological studies provide little support for cells serving as junction “detectors”, many neural models of figure-ground segregation, such as Finkel & Sajda (1992, Neural Computation), determine occlusion by assuming such cells exist. Border-ownership cells identified in physiological studies by Zhou et al. (2000, Journal of Neuroscience) demonstrate side-of-figure and occlusion sensitivities even at non-junction contours, thereby integrating global grouping information, presumably through feedback. We present a biologically plausible neural model of primate visual areas V1 and V2 that performs figure-ground segregation using grouping circuits without dedicated junction mechanisms. Our model LGN's on-center/off-surround competition is followed by oriented contrast detection by V1 complex cells. We use border-ownership cells that receive feedback from grouping cells that respond to convexity, similar to those of Craft et al. (2007, Journal of Neurophysiology). We assume these cells assert “figureness”, but they inherently face a static aperture problem. For instance, individual grouping cells can only resolve T-junctions as two L-junctions, which do not signal occlusion. We therefore use another layer of grouping cells that, through competitive feedback across scale, can correctly vote for occlusion using a winner-take-all network, consistent with human judgments of synthetic and natural T-junctions presented through an aperture in McDermott (2004, Perception). Our model also explains why humans report occlusion in one but not both of the E-shaped occluders of Rubin (2001, Perception), and in more general visual displays, such as the Kanizsa square and transparent regions. This suggests that the visual system may not rely on specialized junction circuits to perform figure-ground segregation. Occlusion information conferred by T-junctions may reflect a more general process of inter-scale competition of grouping cells sensitive to local convex regions.
This PDF is available to Subscribers Only