Image processing applications commonly assume a relatively transparent
transmission medium, unaltered by the atmospheric conditions. Outdoor vision
applications such as surveillance systems, intelligent vehicles, satellite imaging,
or outdoor object recognition systems need optimal visibility conditions in order
to detect and process extracted features in a reliable fashion. Since haze
degradation effects depend on the distance, as disclosed by previous studies [1,
2] and observed as well in our experiments (see Fig. 1), standard contrast enhancement
filters such as histogram stretching and equalization, linear mapping,
or gamma correction are limited to perform the required task introducing halos
artifacts and distorting the color. The contrast degradation of a hazy image is both multiplicative and additive. Practically, the haze effect is described by
two unknown components: the airlight contribution and the direct attenuation
related to the surface radiance. The color ambiguity of the radiance is due to
the additive airlight, which increases exponentially with the distance. Enhancing
the visibility of such images is not a trivial task, as it poses an inherently underconstrained
problem. A reliable restoration requires an accurate estimation of
both the true colors of the scene and the transmission map, closely related to
the depth-map.