An improved SIFT-based algorithm for SAR images registration
Haolong Chen, Zi Wang, Tao Sun, Zhizhou Chen, Ziyi Zhou, Zegang Ding · IET conference proceedings. · 2026
Synthetic Aperture Radar (SAR) image registration faces challenges in matching accuracy due to multiplicative speckle noise and geometric distortions. To address this, this paper proposes an enhanced SAR-SIFT registration method with three key improvements: First, we implement image enhancement through a Median-LEE hybrid filter leveraging local statistical characteristics, achieving significant speckle noise suppression. Subsequently, during feature point detection, we replace fixed gradient magnitude thresholds with an adaptive dynamic threshold that self-adjusts based on localized gradient statistics (mean and maximum values within sampling windows), effectively minimizing false detections. Finally, for feature matching, we introduce a Bidirectional Nearest Neighbor Distance Ratio (Bidirectional NNDR) approach, integrated with RANSAC-based mismatch removal. This strategy yields significantly higher correct matching rates compared to unidirectional methods. Experimental results confirm that the enhanced SAR-SIFT algorithm demonstrates marked improvements over the original method, notably reducing false detections, increasing correct matching rates, and improving computational efficiency.