Color and sharpness assessment of single image dehazing

Image dehazing is the process of enhancing a color image of a natural scene that contains an undesirable veil of fog for visualization or as a pre-processing step for computer vision systems. In this work, we investigate the performances of eleven state-of-the-art image quality metrics in evaluating dehazed images, and discuss challenges in designing an efficient dehazing evaluation metric. This is done through a composite study based on the agreement between subjective and objective evaluations. Accordingly, we evaluate five state-of-the-art dehazing algorithms. We use two semi-indoor scenes, degraded with several levels of fog. One important aspect of these scenes is that the fog-free images are available and can therefore serve as ground-truth data for dehazing methods evaluation. This study shows that the best working dehazing method depends on the density of fog. There seems to be a clear distinction between what people perceive as good quality in terms of color restoration and in terms of sharpness restoration. Most metrics show limitations in providing proper quality prediction of dehazing. According to the introduction and analysis, a contribution of this work is to point out the flaws in the evaluation and development of dehazing methods. Our observations might be considered when designing efficient methods and metrics dedicated to image dehazing.