Noise fingerprint-based infrared fixed pattern noise removal

Fixed pattern noise (FPN) is a common artifact in digital infrared sensors, caused by inherent manufacturing imperfections, and often leads to severe image quality degradation. In this paper, we propose a novel, to our knowledge, method for FPN removal based on camera noise fingerprints, termed noise fingerprint guided denoising network (NFGD-Net). The proposed framework integrates a main denoising network equipped with channel-wise attention blocks and a noise fingerprint extraction subnetwork. The extraction module adopts a Siamese architecture to capture noise residuals, which are further enhanced using a Haar discrete wavelet transform-based attention mechanism to extract directional noise features. Additionally, a gated feature modulator is introduced to improve the network's feature learning capability. By leveraging the structured characteristics of camera-specific noise fingerprints, the proposed method effectively suppresses FPN and restores high-quality infrared images. Extensive experiments on two infrared image datasets demonstrate that NFGD-Net consistently outperforms state-of-the-art methods in both qualitative and quantitative evaluations. Notably, under uniform-distributed noise with an intensity of 0.05, our method achieves a PSNR of 41.61 and an SSIM of 0.9862, showcasing its strong ability to preserve fine details while eliminating structured noise.