Improved bidirectional image registration based on Radon-SIFT
Fan Bu, Yuehong Qiu, Jinxia Liu, Xingtao Yan · Institutional Repository of Xi'an Institute of Optics and Fine Mechanics, Chinese Academy of Sciences (Xian Institute of Optics and Precision Mechanics) · 2012
Aiming at the problems of large calculation, high complexity and long time-consuming in Scale Invariant Feature Transform (SIFT) feature matching algorithm, this paper proposes a novel bidirectional Radon-SIFT matching algorithm. Based on Radon transform, we make 36 beelines on different directions in local keypoints region. Along these 36 directions, we calculate Radon transform integral values which can be chosen as keypoints vector descriptors. 36-dimensional Radon-SIFT descriptors outweigh the standard 128-dimensional ones both in accuracy and efficiency. The other contribution of this paper is bidirectional matching algorithm. This improved algorithm introduces matching uniqueness constraint to further reduce matching error. To demonstrate the effectiveness and robustness of the proposed algorithm, we apply it to natural images. Experimental results show greater accuracy and faster matching.