Enhanced Underwater Image Restoration: Optimizing Computational Efficiency with U-Net++ and CNN Architectures

Restoring underwater images poses a significant computational challenge, often requiring more time and resources. U-Net and its variants are not much applied for the task of underwater image restoration so we propose a novel approach to make use of the U-net++ architecture. The aim here is to develop a U-Net++ model and a CNN model and analyze the quality of the restored images so they are capable of operating efficiently on systems with limited computational resources while significantly reducing the time required for image restoration. The dataset used for the task of underwater image restoration task in this paper is Large Scale Underwater Images (LSUI). This particular U-Net++ is computationally efficient in contrast to other state of the art models that were developed so far and this helps in saving a lot of time and speed up the process of restoring such images which makes it easier for research and analysis in underwater scenarios. From the results obtained, we conclude that the SSIM 0.9994 and PSNR 70.2617 dB values are higher for the U-net++ model and this architecture can be used for conducting research in underwater environments.