Feature Correspondence with Even Distribution

This paper presents a new method for finding feature correspondences between a pair of stereo images, which can be used to perform 3D reconstruction and object recognition. This paper uses epiploar geometry to determine the location of potential matching points which allow the finding of the correspondences faster and more accurately. An adaptive smoothness algorithm is also proposed to filter out false matches based on the disparity jump in neighboring correspondences. More correspondences are then identified in such a way that they are evenly distributed in the images. Experimental results show that the proposed method effectively improve the percentage of correct matches, total number of correct matches, and even distribution of the correspondences.