Purchase this article with an account.
David Hunter, Paul Hibbrad; The independent components of binocular images reflect the spatial distribution of horizontal and vertical disparities. Journal of Vision 2016;16(12):243. doi: 10.1167/16.12.243.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
The geometry of binocular vision is such that the expected distributions of horizontal and vertical disparities vary from one location in the image to another. Firstly, binocular fixation means that disparities will be smallest in the centre of the image. Secondly, in the natural environment there is a statistical relationship between the height in the visual field and distance, such points lower in the retinal image tend to be closer to the observer. From this, we expect more crossed horizontal disparities in the lower half of the image. Finally, this height-in-the-field effect, combined with the origin of vertical disparities in differential perspective, creates a predictable distribution of vertical disparities. Specifically, we expect points to be shifted downwards in the left eye in the bottom left quadrant of the image, and upwards in the left eye in the bottom right quadrant. We tested the influence of these effects on the disparity tuning of independent components learned from natural binocular images. 139 binocular image pairs were divided into 4 quadrants and 3 eccentricities, and ICA was performed independently for each of the resulting 12 regions. Position disparity tuning of the resulting components was calculated by measuring the shift in their receptive field locations in the two eyes. The distribution of horizontal disparity tunings became broader with increasing disparity, there was a bias towards crossed disparities in the lower half of the image, and towards uncrossed disparities in the upper half. In the lower half of the image, vertical disparities tended to code for lower positions in the left eye in the left half of the image, and for higher positions in the left eye in the right of the image. The distributions of positional disparity tunings learned through ICA therefore reflect the expected distributions of disparities in natural images.
Meeting abstract presented at VSS 2016
This PDF is available to Subscribers Only