Comunicació presentada a: 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, celebrada a Salt Lake City, Estats Units d'Amèrica, del 18 al 23 de juny de 2018.

Image dehazing deals with the removal of undesired lossof visibility in outdoor images due to the presence of fog.Retinex is a color vision model mimicking the ability ofthe Human Visual System to robustly discount varying illu-minations when observing a scene under different spectrallighting conditions. Retinex has been widely explored inthe computer vision literature for image enhancement andother related tasks. While these two problems are appar-ently unrelated, the goal of this work is to show that theycan be connected by a simple linear relationship. Specif-ically, most Retinex-based algorithms have the character-istic feature of always increasing image brightness, whichturns them into ideal candidates for effective image dehaz-ing by directly applying Retinex to a hazy imagewhose in-tensities have been inverted. In this paper, we give theoret-ical proof that Retinex on inverted intensities is a solutionto the image dehazing problem. Comprehensive qualitativeand quantitative results indicate that several classical andmodern implementations of Retinex can be transformed intocompeting image dehazing algorithms performing on pairwith more complex fog removal methods, and can overcomesome of the main challenges associated with this problem.

JVC was supported by the Spanish government grant ref.IJCI-2014-19516, and MB by European Research Coun-cil, Starting Grant ref. 306337, by the Spanish governmentgrant ref. TIN2015-71537-P, & by Icrea Academia Award.