Physics-based Underwater Image Enhancement

Underwater images suffer from serious degradation because of the complex imaging environment. However, the existing methods ignore the degradation caused by the physical factors. It is helpful to improve underwater images via estimate these factors. In this paper, we propose a generative approach for underwater image enhancement, called Physicsbased WaterGAN (PWGAN), which considers the factors that affect the visibility of underwater objects in reality, including the scattering coefficient of different degrees of purity water and the depth. This paper focuses on exploring the scattering coefficient and depth information contained in the degraded and clean underwater images. In our network, the generated results of the two GANs are fine-tuned using the underwater scattering coefficient obtained from the underwater image/the depth information obtained from the clean image to make the results realistic. Experiments show that the PWGAN model has good generalization ability, can adapt to different underwater datasets, and eliminates color degradation of degraded images taken in various underwater environments.