A new image matching method based on combining global and local features

Abstract

This paper is mainly introduced a new approach which is based on combining global and local features. The cascade matching algorithm contains two stages: the first is fast matching regions using global region features extracted from the segmentation regions and the second is matching the Affine-SIFT descriptors from the matched region pairs. In the later stage, RANdom SAmple Consensus (RANSAC) is used to filter some false matches and can gain a large increase in speed. Experiments show the effectiveness and superiority of the proposed method in comparing to SIFT and affine SIFT. The method is also fit for the cases of large scale and orientation changes.