Comparison of local descriptors for automatic remote sensing image registration

Optical remote sensing (RS) images captured in different conditions might exhibit nonlinear changes. The registration of theses image is an important process. In this paper, we evaluate the performance of the three most successful state-of-the-art descriptors in a feature-based registration process. We have separated the detector from the descriptor as their performance depends on the position of the detected features. The descriptors are compared according to their Recall and runtime efficiency and these deals with several geometric and photometric changes. We also proposed an optimization to the SURF algorithm for color images, called O-SURF, which is a combination of the MSER detector and the SURF descriptor. The results show the effectiveness of proposed improvements compared to base SURF version. Finally, based on the test results, we propose an approach to register automatically optical RS images with subpixel accuracy.