Infrared and Visible Image Registration Method Based on Fine-Grained Feature Capture and Multi-Feature Collaboration

In the realm of computer vision, the registration of infrared and visible images is pivotal, harnessing the strengths of distinct imaging modalities to surmount challenges in low-light or adverse weather conditions. Given the intrinsic disparities in imaging mechanisms and the diversity of capture conditions, the development of efficient and accurate registration algorithms is essential. Here we introduce a novel approach to infrared and visible image registration that leverages fine-grained feature capture and multi-feature collaboration, enhancing registration accuracy and speed across various scenarios to meet practical application needs. By employing a multi-branch convolutional neural network (CNN), our method strengthens the extraction of both local and global features. Our main findings reveal that the fine-grained feature capture mechanism and multi-feature synergy framework significantly improve the precision of feature matching and bolster the robustness of the registration process under varying conditions. Furthermore, the composite loss function we propose optimizes registration performance by harmoniously aligning feature consistency, structural similarity, and gradient alignment. This study not only advances image registration technology but also offers new perspectives for multimodal image fusion, with significant implications for applications such as device detection, environmental monitoring, and military reconnaissance.