Abstract
The purpose of this paper is to construct a disparity _??_ection method based on the spatial characteristic of simple cells in the visual cortex and to evaluate its availabilities for the stereopsis. The receptive fields of simple cells are approximated by two-dimensional spatial filters described with Gabor function. This filter is referred to as Gabor filter, which selectively responds of to stimulus patterns with specific spatial frequency and phase.
The principle of the disparity detection is as follows. When input images are given, a pair of the Gabor filters which have a certain phase difference show the maximum output among the filters. The phase difference between the right and left Gabor filters is approximately proportional to the disparity.
The disparity is calculated in the following three steps. First, the left and the right images are spatially filtered with a Gabor filter and the value of phase differences is calculated at each pixel. Second, the phase differences are calculated at several spatial frequencies of Gabor filter. Finally, a unique disparity at each pixel is obtained from a weighted sum of these phase differences, and spatial profile of the disparities in adjacent pixels is made smooth.
