DepthCue: Restoration of Underwater Images Using Monocular Depth as a Clue

In this paper, we perform restoration of underwater images by considering principles of the image formation model in deep neural networks. Typically, underwater images suffer from blur, color loss and other degradations due to the scattering and absorption of light in water as a medium. Quality of restoration is sensitive to depth as scattering and absorption of light increases with depth and introduces a considerable amount of degradation. However, from literature we infer, recent restoration frameworks do not consider the influence of depth on restoration of under-water images. Towards this, we propose to consider depth as a clue for restoration considering relative distance of objects in the scene. We introduce depth with different scales as a clue for learning restoration and term the proposed architecture as DepthCue. We foresee to facilitate the restoration by eliminating the effect of degradations like lost color, blur and noise. We demonstrate our results on benchmark datasets and compare with the state-of-the-art restoration techniques using various quality metrics.