A Self-Attention Guided Approach for Advanced Underwater Image Super-Resolution With Depth Awareness

Underwater image super-resolution (SR) remains a challenging task due to the unique distortions introduced by the aquatic environment, including color fading, scattering effects, and low contrast. In this paper, we propose a novel generative deep learning-based approach to enhance underwater image restoration by integrating Depth-Aware and Self-Attention mechanisms, these mechanisms are an integral part of our models architectural blocks. Unlike adversarial-based methods, this approach is purely generative and focuses on restoring fine image details without using a discriminator. Our method effectively captures multiscale features and refines image details, addressing both local and global distortions that typically hinder SR in underwater images. We incorporate attention gates with our Guided Upsample Blocks (GUB) to improve feature enhancement during the upsampling phase, ensuring that essential image details are preserved. To mitigate issues such as color shifts and low contrast, we incorporate Self Aware Enhancement Block (SAED) in the mid-layers of the network, boosting perceptual quality and visual consistency. Extensive experiments on benchmark underwater datasets demonstrate that the proposed generative method outperforms state-of-the-art techniques across multiple evaluation metrics, including PSNR, SSIM, and UIQM. The proposed architecture provides a robust and effective solution for enhancing underwater image SR, with qualitative and quantitative improvements across various underwater imaging challenges.