Dehazing with Recovery Level Map: Suppressing Over-Enhancement and Residual Haze
Iwamoto Kentaro, Hiromi Yoshida, Youji Iiguni · 2019
Haze degrades contrast and visibility of images, thus it causes bad visibility or poor accuracy in computer vision applications. There are many dehazing methods: prior-based and data-driven methods. Prior-based methods tend to cause over-enhancement such as visual artifacts in the white regions. Data-driven methods cannot sometimes remove haze in the foreground completely. In this paper, we propose a method to suppress both over-enhancement and residual haze based on the dark channel prior (DCP). We use the clarity map as a texture feature and define the recovery level map that determines the amount of dehazing level. We use both the DCP and the recovery level map to estimate the scene transmission. As a result, our method suppresses both over-enhancement and residual haze compared with state-of-the-art dehazing methods.