Identifying an important number of correct corresponding points in images of the same scene/object is an essential task of many computer vision applications. The matching problem is based on finding many reliable corresponding points. A considerable research has been invested in finding ways to describe local feature points as distinctive as possible.
In this work we propose a simple method to increase the number of correct correspondences extracted from a pair of images. The features are extracted and described with the original SIFT (Scale Invariant Feature Transform) method. One of the main reasons is that SIFT has been proven to be the most reliable solution in the matching problems. Even if in general the detected DoG (Difference of Gaussian) feature points have a good repeatability score an important part of them are not found as good correspondences in the original matching procedure.
In this paper we try to improve this lack by considering the color information as additional information in the process of matching. The proposed algorithm calculates the cross correlation and the histogram intersection between neighbour keypoints regions that has been warped based on the estimated scale and orientation between feature patches. Our experimental results show that the new method performs better than the original matching procedure.