Michael Gray, Alexandre Pouget, Richard Zemel, Steven Nowlan, Terrence J. Sejnowski
Local disparity information is often sparse and noisy, which creates two conflicting demands when estimating disparity in an image re(cid:173) gion: the need to spatially average to get an accurate estimate, and the problem of not averaging over discontinuities. We have devel(cid:173) oped a network model of disparity estimation based on disparity(cid:173) selective neurons, such as those found in the early stages of process(cid:173) ing in visual cortex. The model can accurately estimate multiple disparities in a region, which may be caused by transparency or oc(cid:173) clusion, in real images and random-dot stereograms. The use of a selection mechanism to selectively integrate reliable local disparity estimates results in superior performance compared to standard back-propagation and cross-correlation approaches. In addition, the representations learned with this selection mechanism are con(cid:173) sistent with recent neurophysiological results of von der Heydt, Zhou, Friedman, and Poggio  for cells in cortical visual area V2. Combining multi-scale biologically-plausible image processing with the power of the mixture-of-experts learning algorithm represents a promising approach that yields both high performance and new insights into visual system function.
Selective Integration: A Model for Disparity Estimation