SFIVReg: Style Filtering-Based Infrared-Visible Image Registration for UAV Systems
Shuang Su, Ruitao Lu, Yongxiang Yao, Xiangtao Zheng, Yukai Shi, Bin Tang, T B Zhang, Xiaogang Yang · IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 2026
With the widespread application of multisensors in the field of unmanned aerial vehicles (UAVs) remote sensing, the registration of infrared-visible image has become a key step to enhance multimodal remote sensing perception capabilities. However, achieving high-precision registration of UAV infrared and visible images remains challenging due to significant modality differences and complex deformations. In this study, we propose a style filtering-based infrared-visible image registration network (SFIVReg). The core idea of SFIVReg is to design a feature style filtering module to mitigate the style differences between images of different modalities, thereby effectively enhancing the features beneficial for registration. In addition, we design a shared coordinate attention (SCA) fusion decoder, which introduces a SCA mechanism. This mechanism helps the network focus more on locally complex deformations and emphasizes detailed texture features. As a result, the subsequently predicted deformation field becomes more accurate. Finally, the proposed edge-texture alignment loss function guides the model to enhance the constraints on complex edge deformations from the perspectives of edge similarity and deformation smoothness, achieving more precise pixel-level image registration. Qualitative and quantitative experimental results on the constructed dataset demonstrate that the proposed registration method achieves significant performance gains, confirming its effectiveness and stability for multimodal nonrigid registration in complex UAV scenarios.