Efficient feature matching algorithms applied to robust image processing

Abstract

Image feature matching is an essential problem in computer vision. Also, feature matching is an important task in many computer vision applications, such as in VR technology, drones, and autonomous driving. In this paper, four algorithms for matching are presented first. We perform outdoor scene feature matching experiments on four algorithms: Surf, Sift, Orb, Brisk. The original image is rotated, the sharpness is continuously blurred, the scene is transformed, specific regions are scaled, and the resolution is changed. The number of feature points that correctly matched the transformed image to the original image was compared. In this paper, the stability is determined by comparing the operation time of the four algorithms in the case of unified processing. Then, the RANSAC algorithm detects the proportion of internal and external points and compares the robustness. It is concluded that the ORB algorithm is the fastest, and the SURF algorithm is the most robust.