Image matching based on the adaptive redundant keypoint elimination method in the SIFT algorithm
- PDF / 3,465,955 Bytes
- 15 Pages / 595.276 x 790.866 pts Page_size
- 40 Downloads / 203 Views
THEORETICAL ADVANCES
Image matching based on the adaptive redundant keypoint elimination method in the SIFT algorithm Zahra Hossein‑Nejad1 · Hamed Agahi1 · Azar Mahmoodzadeh1 Received: 17 April 2020 / Accepted: 29 October 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Scale invariant feature transform (SIFT) is one of the most effective techniques in image matching applications. However, it has a main drawback: existing numerous redundant keypoints located very close to each other in the image. These redundant keypoints increase the computational complexity while they decrease the image matching performance. Redundant keypoint elimination method (RKEM)–SIFT are incorporated to eliminate these points by comparing their distances with a fixed experimental threshold value. However, this value has a great impact on the matching results. In this paper, an adaptive RKEM is presented which considers type of the images and distortion thereof, while adjusting the threshold value. Moreover, this value is found separately for the reference and sensed images. In an image, the adaptive RKEM finds the histogram of the keypoints distances, for which the number and the width of the bins are determined based on the number of keypoints and the distances distribution metrics. Then, a maximum value for searching the optimal threshold value is determined. Finally, for each integer value smaller than the mentioned maximum, a set containing distances smaller than that value is created and the one with the smallest variance is selected. The integer value corresponding to that set is chosen as the adaptive threshold for that image. This approach can improve the efficiency of the RKEM-SIFT in eliminating redundant keypoints. Simulation results validated that the proposed method outperforms the SIFT, A 2 SIFT and RKEM-SIFT in terms of the matching performance indices. Keywords Keypoint · RKEM · Redundancy index · Image matching · SIFT
1 Introduction Matching is the process of determining correspondence between images of the same scene received at different imaging conditions [1]. This process is an essential step in many applications such as image registration [2], change detection [3], image fusion [4], image mosaicking [5], and 3D reconstruction [6]. Generally, matching techniques can be categorized into 2 types: the area-based and the featurebased approaches [7]. Area-based methods use the distribution of grey-levels, in which images in the windows with the same dimensions are considered directly. To this end, using some similarity (or difference) intensity-based criteria, 2 images are statistically compared in order to determine the location of maximum similarities, representing * Hamed Agahi [email protected] 1
Department of Electrical Engineering, Shiraz Branch, Islamic Azad University, Shiraz, Iran
similar situations [8]. Although the area-based methods can reach very high accuracies of matching, their performance is reduced in areas with uniform textures [9]. Area-based methods are appropri
Data Loading...
