Enhancing Underwater Image Using Multi-scale Generative Adversarial Networks.

Wavelength-dependent light absorption and scattering will reduce the quality of underwater images. Therefore, the characteristics of underwater images are different from those taken in natural. Low-quality underwater images affect the accuracy of pattern recognition, visual understanding, and key feature extraction in underwater scenes. In this paper, we enhance the underwater image using a multi-scale generative adversarial network with adjacent scale feature addition. Adjacent scale feature addition allows the network to more effectively capture the relevant characteristics between two image domains. The multi-scale discriminator can let the enhanced image more closer to the natural image. Our method does not rely on transmission maps and atmospheric light estimation. Experiments on a large amount of synthetic data and real data show that our method is superior to the state-of-the-art methods.