Relaxation Technique for Stereo Images Matching

S.V. Mikhlyaev (Russia)


Image matching, corner detector, gradient operator, relaxation.


This paper develops the approach to image matching based on two stages – preliminary extraction of special points followed by their comparison through the relaxation procedure. Special points (SP), or “points of interest”, are image points that have considerable two dimensional variations of intensity over their neighborhood. An algorithm of subpixel localization of SP of the images formed by local gradient operators such as Harries and Brady corner detectors is considered. The modified relaxation procedure for searching for conjugate points in stereo images is proposed. The procedure is applied to the arrays of SP formed by gradient operators. The strategy for point updating takes into account the angular distribution of points over their neighborhoods along with the strength and unambiguity of the match. The experimental results obtained by using the test images with random points and with points corresponding to computer simulation of stereo images are presented. Also we have presented the results obtained by using the real images. This work was supported by the Russian Foundation for Basic Research, projects no. 00-15-99092 and no. 00 01-00884a.

Important Links:

Go Back