Comparison of image matching algorithms on satellite images taken in different seasons

Image matching, which aims to find the corresponding points in different images, is an important process which is used in various vision-based applications in military, industrial, remote sensing and security systems. Some applications require accurate matching across images taken at different times of the year to be reliable and reusable. Although many detection and description methods are used for image matching, it is important to correctly determine the most robust method for such changes. In this paper we investigate combination of SIFT (Scale Invariant Feature Transform), SURF (Speed Up Robust Features), KAZE, BRISK (Binary Robust Invariant Scalable), FAST (Features from Accelerated Segment Test) algorithms using satellite images that are taken at different times of the year in various seasons and weather conditions. Incorrect matches in the test results are eliminated by MLESAC (Maximum Likelihood Estimation SAmple and Consensus) method. As a result of these eliminations, the accuracy, propagation, changes in the number of the keypoints and the speed of detection of the keypoints are observed. At the end of these analyses, it is concluded that most reliable method in keypoint matching is FAST-SIFT despite the high cost of its computation time