A New Image Matching Algorithm Based on Scale Adapted Interest Point Detection

In this paper, we study the problem of extracting interest points from images and the interest points can be used to perform reliable matching between different scenes. The image matching focuses on two key points: interest point detection and feature matching. However, due to the disadvantage brought by image scale, rotation and view change, the ability of feature matching falls off signally. This paper proposes a new image matching algorithm, which has good performances on image scale, rotation and view change. First, we bring forward a novel interest point detector called scale adapted Harris detector by selecting interest points at different levels of scales. Then we formulate the feature matching based on the Earth Mover's Distance (EMD). At the same time, we implement our proposed algorithm in a benchmark dataset. The experimental results demonstrate the proposed algorithm perform better than other methods in both recall and precision.