The sterometric maching using Gabor Functions versus structured light-shape detection methods

摘要

The most important problem in stereometry is to establish a correspondence between the points on two images representing different views on the same scene. We have developed an algorithm which avoids some disadvantages of previous approaches to this problem. Our approach was based on matching of correlation of some regions from both images. We use a Gabor representation of the images. We define regions on both images and assign the vectors of coefficients of Gabor functions to all those regions. Those vectors are used for comparing of the regions. We adjust an accuracy of matching using local phase information from the coefficients. We also measure the diameter of correlation peek near the region and we increase the diameter of the region if it is smaller than this diameter. In this way we avoid a problem of large areas of the same texture. We also propose implementation of this algorithm in a real time system. The optical method is suggested for computing coefficients of Gabor functions in that system.