Purchase this article with an account.
Jian Ding, Dennis M. Levi; A comprehensive depth perception model with filter/cross-correlation/filter (F-CC-F) structure. Journal of Vision 2019;19(10):263a. doi: https://doi.org/10.1167/19.10.263a.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
We have developed a new depth perception model with a filter/cross-correlation/filter (F-CC-F) structure, and validated it over a broad range of spatiotemporal conditions. The two eyes’ images pass through first-stage spatiotemporal filters to compute the normalized cross-correlation (CC), which goes through a maximum (MAX) operator for solving the correspondence problem, then goes through a disparity window to compute local depth quantities, and finally through a second-stage of spatiotemporal filtering to give the perceived depth. Previous work (Ding & Levi VSS 2016) revealed two normalization mechanisms: (1) an energy model (cross-correlation is normalized by monocular contrast energies) to predict Dmin and (2) a gain-control model (interocular contrast gain-controls before cross-correlation) to predict Dmax. Here, we have developed a full model with multiple pathways to explain both Dmin and Dmax simultaneously. To test the model, we performed rating-scale experiments to measure disparity sensitivities over the whole range of binocular disparities. Stimuli were random-Gabor-patch (RGP) stereograms, consisting of vertical Gabor patches with random positions and phases, but with a fixed spatial frequency (3.0 cpd). We tested five stereogram profiles (circles and annuli of 1.76 – 7.04 arcdeg radius) and eight durations (13 – 2013 ms), revealing the complex spatiotemporal properties of depth perception. We found that disparity sensitivity depends on the stereogram size, which can be explained by a band-pass DOG (Difference of Gaussian) filter as the second-stage filter. The model has four pathways, two for crossed- and two for uncrossed-disparities. Each pathway has its own normalization mechanism (either Energy or Gain-control model), its own disparity window (a disparity sensitivity function, the product of disparity power and exponential decay functions, which increases with stimulus disparity at small disparities but decreases at large disparities), and its own second-stage filter. These second-stage filters were validated experimentally. This comprehensive new F-CC-F model successfully predicts the entire data-set.
This PDF is available to Subscribers Only