iDehaze: Supervised Underwater Image Enhancement and Dehazing via Physically Accurate Photorealistic Simulations

Underwater image enhancement and turbidity removal (dehazing) is a very challenging problem, not only due to the sheer variety of environments where it applicable, but also lack high-resolution, labelled data. In this paper, we present novel, two-step deep learning approach for underwater dehazing colour correction. iDehaze, leverage computer graphics physically model light propagation in conditions. Specifically, construct three-dimensional, photorealistic simulation environments, use them gather large supervised training dataset. We then train convolutional neural network remove haze these images, second transform space dehazed images onto target domain. Experiments demonstrate that our iDehaze method substantially more effective at producing high-quality achieving state-of-the-art performance on multiple datasets. Code, data benchmarks will be open sourced.