Self-Attention Underwater Image Enhancement by Data Augmentation

Underwater optical image play a significant role in the exploration ocean. However there are diverse causes of distorted underwater optical scenes, such as light refraction, absorption, scattering and so on. Images enhancement is indispensable in low-level and high-level underwater vision task. Therefore, a novel method based on Generative Adversarial Networks is presented in this paper, which is able to recover lost information for underwater distorted images. No matter from color, detail or texture, the underwater images reconstructed by our approach been greatly improved. In addition, a novel solution is provided for lacking paired dataset which is needed for network train. At last, many qualitative and quantitative experiments are carried out, which can demonstrate that our approach proposed in this paper is robust and effective.