Optimization of stereo disparity estimation using the instantaneous frequency

Abstract

The use of phase differences from Gabor filter responses is a well established technique for the computation of stereo disparity. It achieves the subpixel estimation of disparity without applying a correspondence search. However, the problem of scale or central frequency selection is still unsolved. Here, we study the effects of varying filter frequency on the disparity estimation and we compare it to the use of the instantaneous frequency. The analytical results on several models of intensity and disparity variation show robustness of the disparity estimates against variations in the filter-wavelength.