Survey on Underwater Image Enhancement using Deep Learning

Underwater image quality improvement is very much needed in underwater imagery as images obtained may contain huge amounts of noise, haziness, and color loss. Several steps are done to remove these and obtain high-resolution images. One of the main techniques used for processing is by using General Adversarial Networks(GANs) giving super resolute images as result. Basic traditional methods are the usage of filters and deep learning methods which are the most commonly used techniques. In this paper, different types of GANs are discussed and their evaluation is studied. The GANs studied include WaterGAN, Pyramid Attention Mechanism-Oriented Symmetry Generative Adversarial Network (PAMSGAN), Underwater GAN(UWGAN), SpiralGAN and General Adversarial Networks used to improve LiDAR (Light Detection and Ranging) images and real-time underwater images which in result show great results when compared to traditional methods of underwater image enhancement.