Colorimetric Screening of the Haze Model Limits

The haze model, which describes the degradation of atmospheric visibility, is a good approximation for a wide range of weather conditions and several situations. However, it misrepresents the perceived scenes and causes therefore undesirable results on dehazed images at high densities of fog. In this paper, using data from CHIC database, we investigate the possibility to screen the regions of the hazy image, where the haze model inversion is likely to fail in providing perceptually recognized colors. This study is done upon the perceived correlation between the atmospheric light color and the objects' colors at various fog densities. Accordingly, at high densities of fog, the colors are badly recovered and do not match the original fog-free image. At low fog densities, the haze model inversion provides acceptable results for a large panel of colors.