Multispectral thermal imager temperature field matching method based on co-occurrence filter and SIFT

Multispectral thermal imagers offer fast sampling speed, a wide temperature measurement range, and the ability to mitigate emissivity-related errors. Nevertheless, the optical characteristics of the multi-aperture beam-splitting structure and limitations in instrument precision introduce aberrations in captured temperature field images, reducing measurement accuracy. More importantly, since the radiative characteristics of the same object differ significantly across wavelengths, and the output resolution of temperature fields is relatively low under a large field of view, traditional SIFT-based methods struggle to achieve reliable matching in this scenario. To overcome these challenges, we propose a multispectral temperature field matching algorithm based on co-occurrence and Gabor filters combined with SIFT (CoGF-SIFT). The method first applies co-occurrence filtering to enhance low-frequency information, thereby improving the extraction of thermal features such as heat transfer and convection inside high-temperature regions. Then, Gabor filtering is used to enhance high-frequency information, strengthening boundary texture features between high- and low-temperature regions. Finally, multi-level features are fused, and robust keypoint correspondences are obtained through SIFT extraction and RANSAC-based matching. Experimental results demonstrate that the proposed algorithm not only solves the matching failures observed with conventional methods but also extracts 3–5 times more feature points, while reducing the root mean square error (RMSE) by 28 %–61 %. These improvements highlight the robustness and accuracy of CoGF-SIFT in multispectral thermal imaging. The method provides a reproducible solution with broad application potential in industrial inspection, structural health monitoring, medical thermal diagnostics, and environmental sensing.