Position-guided weighted structural feature matching method

This paper proposes a Position-Guided Weighted Structural Feature Matching (P-WSM) method to address the challenges of nonlinear radiometric differences and geometric distortions in multimodal image matching. The method first optimizes keypoint distribution through a feature aggregation strategy, then constructs a weighted structural space (WSS) by integrating Log-Gabor filter responses, Gaussian-smoothed maximum moments, and original images, which effectively enhances feature stability across modalities. Based on this representation, we extract FAST corner points and develop an improved GLOH descriptor for robust feature description. To further improve matching accuracy, we incorporate a Fast Sample Consensus (FSC) algorithm with position-guided k-nearest neighbor search strategy. Experimental validation using LPIPS metrics demonstrates the effectiveness of our approach in improving multimodal image similarity. Comprehensive evaluations on diverse datasets including visible-depth, infrared, map, SAR, and day-night images show that P-WSM achieves significant improvements over the classical WSSF method, with a 9.4% increase in number of correct matches (NCM) and 25.1% reduction in root mean square error (RMSE). These results highlight the method's strong capability in handling complex multimodal matching scenarios with substantial radiometric and geometric variations.