FSIRD: Mining Frequency and Spatial Domain Features for InfRared Image Denoising
Xiaozhong Tong, Shaojing Su, Siyang Huang, Zongqing Zhao, Desen Bu · 2025
Infrared images are widely used in various applications, and infrared image denoising (IRD) is the basis for advanced vision tasks, such as target detection and tracking. In recent years, image-denoising methods have employed deep learning as the main paradigm. However, the development of IRD has been slow owing to the lack of publicly available datasets for IRD. In this study, we construct a large-scale IRD benchmark dataset containing multiple complex backgrounds. In addition, we design a benchmark framework FSIRD for infrared image denoising, which consists of a dual-domain residual feature extraction module focusing on the frequency and spatial domains of the image and a bottleneck for enhanced multi-scale feature representation. The study results demonstrated that the proposed method achieves state-of-the-art performance on the constructed dataset and recovers degraded IR images with high quality. The new IRD dataset is available at https://github.com/aurora-sea/IRAY.