Single image dehazing for real colored-haze images

Abstract

Since the problem of image dehazing is of great significance and challenging, it has attracted many researchers for a long time. Many excellent dehazing algorithms have emerged one after another. However, few works focus on dehazing tasks with colour deviations. Existing methods do not perform as well as expected on real image dehazing, especially on images with colour deviations. The proposed method mainly focuses on this task and achieves competitive performance. In this paper, a two-step network is proposed, which consists of the Colour Balance Module and the Dehazing Module. Hazy images are first fed into the Colour Balance Module to generate colour corrected hazy images, which are then sent to the pre-trained dehazing net to get the real colour corrected haze-free results. A coloured haze image database is also constructed. Experiments and comparisons are implemented to show the effectiveness of the proposed method, not only on real coloured-haze images but also on other dehazing tasks.