Abstract
Image registration has been widely used in image mosaic, medical image processing, robot vision, pattern recognition and other fields. Normally, the similarity measure that measures the similarity between feature point structures is employed to determine whether feature points in the image are matched to those in the reference image. However, this measure cannot guarantee all the feature points to be correctly matched. The existing state-of-the-art technologies like the Scale Invariant Feature Transform (SIFT) or the Speeded-Up Robust Features (SURF) also suffers from mismatched feature point problem, which will degrade the registration accuracy. To remedy this, this paper proposes a novel algorithm based on SIFT and the verification mechanism of feature point pairs. Firstly, the feature points set of initial matches is obtained by the SIFT algorithm. Then the invariance of affine transformation is used to test the corner set and selects accurate reference points for image registration. Finally, the feature points with similar structure are used to carry out the angle constraint. Experimental results show that the proposed verification mechanism can eliminate false matching points with similar structure, and thus improve the accuracy of image registration.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have