Multi-level registration method for visible and infrared images guided by common optical axis priors

The visible-infrared dual-mode co-aperture seeker imaging system exhibits near-co-axial characteristics, providing a priori assumptions regarding center alignment and scale correlation for dual-mode image registration. However, due to differences in imaging mechanisms and sensors, scale, grayscale, and local structural mismatches still exist in the dual-mode images, limiting the stability and accuracy of cross-modal registration. To address these issues, this paper proposes a multi-level registration method for visible-infrared images guided by a common optical axis prior. The method first utilizes the approximate common optical axis property of the co-aperture system to establish initial scale constraints; subsequently, it constructs a phase-congruency-guided structural filtering (PC-GSF) to enhance the consistent representation of edges, contours, and geometric structures across modalities; building on this foundation, a cross-modal structural reliability gradient descriptor (CM-SRG) is designed. Through local structural normalization, reliability-aware gradient encoding, and multi-scale aggregation, the robustness and discriminative power of feature descriptions are enhanced, ultimately achieving stable registration of visible and infrared images. Experimental results show that, in experiments using a typical image pair from the VIR-UAV dataset, compared to the most recent Cross-Scale method among the selected comparison methods, the RMSE, MAE, and MEE of the method proposed in this paper were reduced by approximately 44.0%, 32.4%, and 46.4%, respectively, while the runtime was reduced from 24.47 s to 6.28 s, demonstrating the effectiveness of the proposed method in terms of cross-modal registration accuracy, local error control, and computational efficiency.