Single image dehazing using multiple transmission layer fusion

Abstract

Methods for single image dehazing have been widely studied based on the atmospheric scattering model and dark channel prior (DCP); they usually adopt an additional refinement procedure such as guide filtering to restrain the halo artefacts, but it easily induces undesirable textures in the final transmission map, and further leads to an overall contrast reduction and detail blur. In this paper, an efficient approach was proposed to enhance single hazy images without any refined post-process, which is based on the strategy of multiple transmission layers fusion. In order to estimate the final transmission map adapting to different scenes reasonably, the multiple transmission layers were derived based on DCP with different kinds of adaptive local watch windows. To make sure the atmospheric light is estimated in the most haze-opaque region, the corresponding region was searched hierarchically with the quadtree subdivision method in the top part of the minimal channel of the input image. Finally, the hazy image was restored through solving the scattering model. Comparison experiments verify that the proposed method is straightforward and efficient, which can reduce the halo artefacts significantly, yielding satisfactory contrast and colour for varied hazy images.