Underwater Image Enhancement and Restoration Techniques

Abstract

Underwater images of good quality are required to investigate the underwater environment, which can be used in wide applications such as ocean exploration and underwater object tracking. However, underwater images often have issues, including color casts, color artifacts, and blurred features. The self-attention mechanism in transformers is a more powerful and dynamic option to deal with the shortcomings of CNNs. The ability of transformers to identify and concentrate on vital information is important when dealing with underwater images of varying conditions (water bodies, lighting). Most methods use synthetic underwater datasets, which limit the performance of the data-driven underwater image enhancement or restoration technique. Other approaches are too computationally expensive when dealing with heavy networks and large real underwater image datasets. The main research problem is that there is a need for more accurate underwater image enhancement techniques, to be developed by training and testing on real underwater image datasets. To address the issue as mentioned above, this paper proposes a Window-based Transformer GAN model for underwater image enhancement that generates state-of-the-art performance (PSNR, SSIM, and UIQM) using a collection of underwater image datasets.

Date of Award	Apr 2023
Original language	American English
Supervisor	Sajid Javed (Supervisor)