RKEM: Redundant Keypoint Elimination Method in Image Registration

Abstract

Image features can be identified using keypoints, and there are various algorithms to detect these important points of an image. Scale-invariant feature transform (SIFT) is one of the most applicable algorithms in keypoint extraction. One of the deficiencies of SIFT and its modifications is that they extract redundant keypoints along with the important ones. A keypoint is called redundant if it is very close to another keypoint, and its sum of distances to all other keypoints is smaller. Elimination of a redundant keypoint increases the speed of algorithms in addition to image matching precision. In this study, redundant keypoint elimination method (RKEM) is proposed to remove redundant keypoints of the SIFT algorithm. To do this, the distances between keypoints are calculated. Then if the distance is smaller than a pre-defined threshold, the redundant point is discarded based on the redundancy index. The proposed RKEM method is applied on basic SIFT, and some recent modification of SIFT such as auto-adaptive SIFT and uniform robust SIFT. To investigate the efficiency of the proposed method, a set of experiments is done in image matching and registration. Experimental results show that the proposed RKEM method can improve the efficiency of SIFT family in terms of TP-rate, precision, repeatability, and root-mean-square error.