LOSC: Optical-SAR Image Registration Based on Local Oriented Scale Contour Feature Descriptor

Abstract

To address challenges in multimodal registration of synthetic aperture radar and optical images—including nonlinear radiation differences, multiplicative speckle noise, and large geometric rotations—we propose a robust registration method based on a local orientation scale contour (LOSC) descriptor. First, an orientation coherence gating and harmonic fusion operator is developed to extract common structural features with high signal-to-noise ratio. It introduces a global orientation coherence gating factor to suppress unstructured noise and employs a harmonic averaging mechanism to complementarily fuse multiscale energy and phase congruency, thereby significantly enhancing edge saliency in heterogeneous images. Second, an LOSC descriptor is constructed from the fused features. A coupled transformation scheme combining spatial geometric rotation and orientation channel shifting is designed to produce feature representations that are highly invariant to arbitrary rotations. Finally, to overcome the lack of initial global geometric constraints, we propose a global rotation estimator based on geometric identity isolation. By leveraging a high-dimensional feature space of principal sequence ranking and an improved ratio test strategy, the global rotation angle can be rapidly estimated without any prior calibration. Experimental results show that the proposed method achieves superior correct matching rate, registration accuracy (RMSE), and time efficiency compared with state-of-the-art algorithms, confirming its robustness in complex remote sensing scenarios.