A Comparative Analysis of Feature Detection and Matching Algorithms for Aerial Image Stitching

Abstract

Features detection and matching are the essential processes in image mosaicing and computer vision applications. Our work intend to find descriptors that are obtained by considering all interest/feature points and its locations on images, and then form a set of corresponding spatial relations based on the interest points between images. Hence in this paper, we will evaluate and present the performance of a few detectordescriptor-matcher approaches on raw aerial images for stitching image purposes. We have experimented on Canny Edge Detector, SIFT and SURF approaches to extract feature points. The extracted descriptors are then matched using FLANN based matcher. Finally, the RANSAC Homography is used to estimate the transformation model so stitching procedure could be applied in order to produce a mosaic aerial image. The results have shown that SURF approach outperforms the others in terms of its robustness of the method and higher speed in execution time.