A two-stage approach for single thermal image restoration

Abstract

Thermal images are prone to significant degradation due to noise, low contrast, and loss of fine details, which poses challenges in many practical applications. Traditional image restoration techniques, particularly those developed for the RGB domain, struggle to effectively balance noise reduction, contrast improvement, and detail preservation when applied to thermal images. In this work, a novel two-stage deep learning framework designed to address these issues in thermal image restoration is proposed. The approach separates the task into a denoising stage and a contrast enhancement stage, with a particular emphasis on preserving fine details throughout the process. By employing a detail extraction mechanism, the method ensures that crucial image details are maintained, even as noise is reduced and contrast is enhanced. Extensive experiments demonstrate that the method not only outperforms state-of-the-art techniques in terms of PSNR and SSIM, but also excels in preserving fine details.