Abstract/Summary

Image registration is a crucial task in many computer vision applications. It is the process of matching and aligning two or more images of a scene. These images can be captured from different viewpoints, different sensors, or different times. Feature based image registration has four main steps: feature detection and description, feature matching, outliers rejection and computing homography and image re-sampling. Computational cost and registration accuracy of feature-based image registration mainly depend on the robustness of feature detection and description methods. Therefore, choosing an optimal feature detection and description method is vital in image registration applications. This research illustrates a comparison between popular image registration algorithms; Scale-invariant feature transform (SIFT), Speeded Up Robust Features (SURF), Oriented FAST and Rotated BRIEF (ORB), KAZE, Binary Robust Invariant Scalable Keypoints (BRISK) and Accelerated-KAZE (AKAZE) in different scenarios: rotation (0 to 360 degrees), scaling (25% to 600%) and multitemporal. The remote sensing images that are used in the experiments are Radar images, Aerial images, and Unmanned Aerial Vehicle (UAV) images. Nearest Neighbour Distance Ratio (NNDR) is performed in the feature matching, whereas RANSAC is applied to reject the outliers matching. The results of the experiments show that SIFT outperforms other algorithms, showing strong stability and high precision in all scenarios. As for real-time application, ORB performs well, and it is the fastest algorithm for all scenarios and then AKAZE as the second fastest one