MFDeS: a multi-frame destriping method

Stripe noise is a pervasive non-uniformity characteristic in thermal images captured by focal plane arrays (FPAs), severely impacting vital image processing tasks such as video stabilization, super-resolution, and target tracking. In response to this challenge, researchers have developed three primary categories of methodologies: statistical methods, optimization techniques, and deep-learning approaches. However, most existing destriping methods focus on individual frames, often resulting in overfitting to the stripe noise. To address this limitation, we present a cutting-edge destriping methodology called MFDeS , which harnesses the power of multi-frame analysis. By integrating the gradient features of the stripes with background information derived from consecutive frames, MFDeS offers a more holistic solution to stripe noise. In an extensive evaluation involving both simulated and real datasets, MFDeS consistently outperforms other baseline methods across a range of stripe noise levels and random errors, showcasing significant improvements in both visual clarity and numerical metrics. Additionally, thorough consistency checks and sensitivity analyses reveal MFDeS ’s robustness in real-world applications. Thus, MFDeS not only enhances the theoretical framework for destriping but also marks a substantial advancement in non-uniformity correction for thermal imaging systems equipped with FPA detectors. This innovative approach promises to elevate the precision and reliability of thermal image analysis across various practical scenarios.