A Novel Iterative SIFT Points Matching Method

xiangru li,tan yang,yu lu,zhiheng wang

DOI: https://doi.org/10.1007/978-3-642-54924-3_59

2014-01-01

Abstract:This work focuses on feature point matching problem based on local descriptor scale-invariant feature transform (SIFT). Feature point matching is a fundamental problem in image analysis and computer vision, and SIFT is one of the most typical local descriptors. Essentially, the available methods of SIFT matching treat the feature points equally in matching, which results in very sparse matches. To increase the density of matched SIFT points, we propose a scheme to improve the performance of SIFT matching by mining additional information from the topological relationships between feature points and from the matching procedures. The fundamental idea is that the distinctiveness are different from point to point, some are easy to be matched correctly others are relatively unreliable to be matched directly; and we can discover some supplementary information from matched points to reduce ambiguity of the less distinctive points and improve their matching accuracy. Experiments verify the performance of the proposed method.