AquaGAN: Restoration of Underwater Images

In this paper, we propose a generative model to restore degraded underwater images considering attenuation coefficients as clue and name it as AquaGAN. Computing the attenuation coefficients as given in revised image formation model demands in-situ measurements. However, in-situ measurements in underwater scenario is infeasible. Towards this, we propose to estimate the attenuation coefficients using learning based methods and use these parameters as clue for restoration of degraded underwater images. Restoration of true colors in underwater scenario is challenging as intensity of light changes with distance. Preserving true colors during restoration by minimizing single objective function may affect the quality of restored image. Towards this, we propose to combine different objective functions for restoration of true colors. We demonstrate the results of restoration on benchmark dataset and compare the results of proposed methodology with state-of-the-art methods both qualitatively and quantitatively.