UDC 004.932.2
CORRELATION-BASED METHOD FOR SYMMETRIC REFINEMENT OF KEYPOINT COORDINATES
A. A. Denisov, post-graduate student, RSREU, Ryazan, Russia;
orcid.org/0009-0006-6765-5054, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
A. I. Novikov, Dr. in technical sciences, full professor, RSREU, Ryazan, Russia;
orcid.org/0000-0002-8166-8234, e-mail: This email address is being protected from spambots. You need JavaScript enabled to view it.
In keypoint-based image alignment, not only is the correctness of point matching critical during the cor respondence stage, but also the accuracy of keypoint coordinate localization itself. The analysis of cross correlation values between matched keypoints has shown that conventional detectors often localize keypoints with slight deviations. These localization errors are typically small – around 1 – 2 pixels along one or both image axes; but the resulting differences in patch-wise correlation coefficients between incorrectly and cor rectly located points can range from 0.6 to 0.9. The aim of this study is to demonstrate that popular keypoint detectors may introduce such shifts and to present a method for refining keypoint positions. The proposed approach, a correlation-based symmetric refinement method, evaluates zero-mean normalized cross correlation (ZNCC) across all possible pairs of points within neighborhoods surrounding initially matched keypoints. Unlike conventional methods that refine only one point, the proposed technique simultaneously adjusts both keypoints in a matched pair. The experiments conducted on real-world imagery show that the method yields a substantial improvement in overall matching quality. The proposed method can be applied in computer vision systems where high-precision image matching is critical-such as mapping, aerial imag ing, medical visualization, robotics, and in tasks involving the fusion of images captured from different view points and sensors.
Key words: keypoint, detector, keypoint descriptor, correlation coefficient, keypoint neighborhood anal ysis, matching accuracy.