Enhancing Underwater Images with MultiScale Residual Attention Network

Most underwater images will be distorted because of light scattering and attenuation. Nowadays, the concept of ocean engineering and underwater robotics require high-resolution underwater photos. Underwater image enhancement algorithms based on CNNs and GANS have been developed. Due to a lack of appropriate datasets, many algorithms fail to produce satisfactory results. However, the lack of accuracy and speed makes them less advanced than other image processing methods. The project aims to use a multi-scale residual attention network. It scales the image, extracts local features, and reforms the image. The computational costs will be reduced by adding new features, reducing distortion in the photo, or preprocessing it after the training