Multiscale stereo analysis based on local-color-phase congruency in the color monogenic signal framework

A multiscale method based on local-color-phase congruency in the color monogenic signal framework is proposed to match feature points and estimate disparity maps for stereo images. As the monogenic signal is the extension of the analytic signal to gray-level images using the Dirac operator and the Laplace equation, the scale-space color monogenic signal is the extension of the monogenic signal to color images based on Clifford algebras. The local color phase, which is estimated by computing the geometric product between the color monogenic signal and a unit reference vector in red-green-blue color space, contains the color and geometric structure information and is robust against noise and brightness change. Experimental results on synthetic and natural stereo images show the performance of the proposed approach.