Mismatching points rejection in multi-modal super-wide field-of-view infrared distorted image registration with global selection-preserving matching

Super-wide field-of-view (SWFOV) long-wave infrared (LWIR) image registration has important applications in military reconnaissance, intelligent driving, etc., but its complex geometric distortion, low contrast and noise disturbance lead to insufficient accuracy of traditional registration algorithms. In our work, an improved Speed-Up Robust Feature (SURF) algorithm combined with the proposed Global Selective Preserving Matching (GSPM) strategy is proposed to solve the mismatching point rejection challenge. This algorithm dynamically adjusts feature detector thresholds and density to adaptively extract key matching points. By integrating global geometric consistency filtering and local triangulation verification from the GSPM algorithm, it can dual constraint to reject mismatching points while preserving topological integrity. The experiments show that our method performs significantly better than traditional methods with Matching Precision (MP) close to 1, Root Mean Square Error (RMSE) as low as 0.0457 pixels, and Structural Similarity Index Measure (SSIM) up to 0.947 in six sets of indoor and outdoor scene data. And comparative experiments are conducted to demonstrate the superiority of this algorithm in image registration for SWFOV infrared gazing images with large distortion characteristics. In the low-noise environment, its anti-noise performance is better than that of the comparison algorithm, and the computation time only increases by 0.1–0.3 s, which meets the real-time demand. This research provides a reliable solution for high-precision rejection of mismatching points in SWFOV LWIR images for data fusion in multi-modal imaging systems to address challenges posed by intricate systems in various domains, including healthcare, transportation, environmental monitoring, target detection and beyond.