Self-Supervised Transmission-Guided Network for Underwater Image Enhancement

Underwater image processing is becoming increasingly popular for improving underwater exploration, including tasks such as underwater terrain scanning and autonomous underwater vehicles (AUVs). However, underwater images often suffer from attenuation, color distortion, and noise, which degrade quality and limit application effectiveness. Moreover, collecting accurate ground truth underwater image datasets for training models is a daunting challenge. Therefore, we propose a new end-to-end underwater image enhancement network based on the synthesis of an underwater dataset. Our primary method involves the generation of a paired underwater dataset and the new transmission-guided model for underwater image enhancement. We transform over-land (or in-air) images into underwater images and enhance the information by incorporating the transmission map, thereby creating a dataset for training end-to-end underwater image enhancement models. In addition, we propose a transmission-guided network to improve image quality and enhance details by fusing the underwater images and the transmission maps. Experimental results have shown that our proposed framework outperforms the state-of-the-art methods in the field of underwater image enhancement.