Purchase this article with an account.
Harald Ruda, Ennio Mingolla, Stephen Grossberg, Arash Yazdanbakhsh; Modeling Hyperacuity Data with a Hierarchical Neural Vision Network and Modified Hebbian Learning. Journal of Vision 2013;13(9):276. doi: https://doi.org/10.1167/13.9.276.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Visual hyperacuity enables an observer to make accurate judgements of the relative positions of stimuli when the differences are smaller than the size of a single cone in the fovea. Hyperacuity can be used as a gauge for testing various characteristics of the visual system, and it can also be used as a stringent test for models of the visual system. A variant of the Boundary Contour System (BCS) is here used to address previously unexplained psychophysical hyperacuity results involving contrast polarity, stimulus separation, and sinusoidal masking gratings. Two-dot alignment thresholds were studied by Levi & Waugh (1996), who varied the gap between the dots, with same and opposite contrast polarity with respect to the background, and also with and without a broadband sinusoidal grating of different orientations. They found that when the gap between the dots is small (6 arcmin), different patterns of results are obtained for the same and different contrast polarity conditions. However, when the gap is large (24 arcmin), the same pattern was obtained irrespective of contrast polarity. The simulations presented here replicate these findings, producing the same pattern of results when varying the gap between the dots, with same and opposite contrast polarity with respect to the background, and also with and without a broadband sinusoidal grating of different orientations. The vision model used (BCS) is able to produce these results because it includes stages of contrast insensitivity (complex cells), spatial and orientation competition, and long-range completion. A novel aspect of the model is the use of sampled field processing, which simplifies the model’s description in terms of equations. In addition, modified Hebbian learning and a neural decison module are proposed as mechanisms that link the vision model’s outputs to a decision criterion. All model parts are biologically plausible and have physiological correlates.
Meeting abstract presented at VSS 2013
This PDF is available to Subscribers Only