EF-GLOH: An Efficient Visible and Infrared Image Matching Algorithm Based on Phase Congruency and Gradient Enhancement

The complementary fusion of information from visible and infrared images plays a crucial role in fields such as security surveillance and autonomous driving. Image matching serves as the foundation for the effective fusion of visible and infrared data. However, due to differences in imaging sensors, significant nonlinear radiometric differences exist between the two modalities. Matching algorithms relying on radiometric information often struggles to obtain robust and accurate correspondences, leading to low matching precision and even mismatches. To address this issue, this paper proposes a radiometric invariant image matching algorithm (EF-GLOH) based on phase information and gradient enhancement. In the aspect of feature point detection, a weighted equation leveraging image phase information is designed to construct a feature detection map that more comprehensively captures image information, ensuring sufficient and uniformly distributed feature points in both visible and infrared images. For descriptor construction, an image information equalization method based on guided filtering is proposed. Building upon this, the third-order Sobel gradient of the filtered image is obtained, and an improved Gradient Location and Orientation Histogram (GLOH) is used to describe the feature points. Experiments conducted on public datasets (RGB-NIR Scene, Road Scene, and OSU Color-Thermal) demonstrate that the proposed method achieves precise matching across all three datasets, with an average RMSE of 1.51. Compared with five state-of-the-art methods, including LGHD, OS-SIFT, CAO-C2F, RIFT, and LNIFT, the proposed method achieves the highest number of matches and the best matching accuracy.