Contents 1. INTRODUCTION
The presence of haze or fog can significantly degrade the visibility of an image captured in outdoor environments. Recovering high-quality images from degraded images (a.k.a. image dehazing) is beneficial for many realistic applications, e.g. video surveillance, unmanned vehicles, object recognition and tracking, etc [1]. The atmospheric scattering model can be used to describe the hazy image generation process, i.e., \begin{equation*}{\mathbf{I}}({\text{x}}) = {\mathbf{J}}({\text{x}})t({\text{x}}) + (1 - t({\text{x}})){\mathbf{A}},\tag{1}\end{equation*}
where I is the observed hazy image, J is the haze-free scene radiance to be restored, A is the global atmospheric light, and t is the transmission map related to depth map. The purpose of image dehazing is to recover J from I, which is particularly challenging since both transmission t and atmospheric light A are unknown. Several physically grounded priors, e.g., dark channel prior (DCP) [2], color-lines prior [3], color attenuation prior [4], non-local prior [5], and color ellipsoid prior [6], have been proposed to assist in improving image dehazing. We will mainly consider the DCP-based dehazing methods since other priors fall beyond the focus of this work.
