Abstract
In a computer vision, a binocular stereo vision is a representative method for obtaining the 3-D information of object. The main issue in processing stereo images is to decide the correspondence points between a left image and a right image, that is, a stereo matching problem. Stereo matching on the images including occlusion and reversal position is a difficult problem. Efficient and effective methods have been not explored yet.
In this paper, we propose two new stereo matching algorithms. One is based on synergetics, which H. Haken proposed to describe self-organization in a dynamical system. The other is based on dynamic programming (DP). The stereo matching algorithm using synergetics is expressed with nonlinear differential equations with two parameters, which are called an order parameter and an attention parameter. The order parameter presents the degree of relationship between a feature point on one image and a correspondence candidate on the other image. The attention parameter has a role controlling the growth of order parameter in the differential equations. We incorporate various information on objects into the two parameters. The stereo matching algorithm using DP is a revision of a conventional DP stereo matching.
The two methods are applied to stereo matching experiments on wood block images including occlusion and reversal position. It is shown that the two methods have quite higher matching precision than a block matching method, and the algorithm based on synergetics has the highest matching precision.
