Auto-Adaptive Well-Distributed Scale-Invariant Feature for SAR Images Registration

Registration of SAR images needs well-distributed and reliable matching point pairs，but the traditional Scale-Invariant Feature Transform(SIFT) approach has limitations related to the distribution of extracted features. A well-distributed auto-adaptive SIFT matching algorithm for SAR images is proposed，based on local textures，using a double-threshold-optimization selection strategy.The proposed algorithm auto-adaptively controls the distribution of the extracted feature points and maintained reliable and precise in both the image and scale space. Comprehensive evaluation of the efficiency，distribution quality，and matching accuracy of the extracted point pairs on a variety of SAR images demonstrates the capabilities of the proposed algorithm.