Performance Evaluation Of Well-Known Feature Detectors And Descriptors

The detection and matching of feature points is an important part in many computer vision applications. In this paper, we explore the performance of six state-of-the-art detectors and descriptors which are SIFT with SIFT, SURF with SURF, BRISK with FREAK, BRISK with BRISK, ORB with ORB and FAST with BRISK. We conduct comparisons of invariance against image transformations such as rotation, illumination, blur and viewpoint in terms of Precision and Matching Ratio. We find that the combination of SIFT with SIFT is most robust to rotation, blur and viewpoint changes. We also find that ORB algorithm performs best under various changes in all binary algorithms.