Thermal Infrared Single Image Dehazing and Blind Image Quality Assessment

Fabian Erlenbusch, Constanze Merkt, Bernardo de Oliveira, Alexander Gatter, Friedhelm Schwenker, Ulrich Klauck, Michael Teutsch; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 459-469

Image dehazing is a method to reduce the effects of haze, dust, or fog in images in order to provide a clear view of the observed scene. A large variety of traditional and machine learning-based approaches exists in the literature. However, these approaches mostly consider color images in the visual-optical spectrum. Apparently, the thermal infrared spectrum is much less affected by haze due to its longer wavelength. But atmospheric perturbations during long-range observation can cause image quality degradation in the thermal infrared (TIR) spectrum as well. In this paper, we propose a method to generate synthetic haze for TIR images. Then, we analyze the already existing blind image quality assessment measure Fog Aware Density Evaluator (FADE) for its applicability to the TIR spectrum. We further provide a comprehensive overview of the current state-of-the-art in image dehazing and empirically show that many approaches originally designed for visual-optical images perform surprisingly well when applied to the TIR spectrum. This is shown in experiments conducted on the recently published M3FD dataset.