Removal of haze and noise from a single image

Images of outdoor scenes often contain degradation due to haze, resulting in contrast reduction and color fading. For many reasons one may need to remove these effects. Unfortunately, haze removal is a difficult problem due the inherent ambiguity between the haze and the underlying scene. Furthermore, all images contain some noise due to sensor (measurement) error that can be amplified in the haze removal process if ignored. A number of methods have been proposed for haze removal from images. Existing literature that has also addressed the issue of noise has relied on multiple images either for denoising prior to dehazing or in the dehazing process itself. 3 However, multiple images are not always available. Recent single image approaches, one of the most successful being the “dark channel prior”, have not yet considered the issue of noise. Accordingly, in this paper we propose two methods for removing both haze and noise from a single image. The first approach is to denoise the image prior to dehazing. This serial approach essentially treats haze and noise separately, and so a second approach is proposed to simultaneously denoise and dehaze using an iterative, adaptive, non-parametric regression method. Experimental results for both methods are then compared. Our findings show that when the noise level is precisely known a priori, simply denoising prior to dehazing performs well. When the noise level is not given, latent errors from either “under”-denoising or “over”-denoising can be amplified, and in this situation, the iterative approach can yield superior results.