Disparity computation of corner points by enhancing feature extractors for sparse narrow-baseline stereo matching

In this paper a method is proposed to match corners for stereo correspondence. Corners convey a lot of information, using which they can be matched. The paper investigates the importance of corners as potential points for accurate disparity map generation. Determining correspondence at corners in stereo image pair is known well enough to generate accurate sparse disparity map, at the same time corners are very unstable as a slight shift in view point changes the background of corner. Therefore area and feature based methods are limited to match corner points with a lower accuracy. In the proposed method the feature matching algorithm is enhanced to match corners using the parallax relation of feature points.