Underwater Image Super-Resolution Reconstruction Based on Hybrid Attention Transformer

Limited by low-light conditions underwater and the complex composition of seawater, images captured using optical cameras often suffer from issues such as blurriness and high noise levels. Utilizing super-resolution reconstruction techniques can address the challenge of obtaining high-definition underwater images more affordably through algorithms, which is crucial for underwater terrain surveys, marine biology research, underwater salvage operations, and more. The Transformer has demonstrated admirable performance in super-resolution reconstruction tasks. However, constrained by network architecture, previous work has been limited to utilizing only a finite amount of spatial information. In order to explore the potential of the Transformer in underwater image super-resolution reconstruction tasks, we propose an approach based on improved hybrid attention transformer (IHAT). It combines the advantages of channel attention and window-based self-attention mechanisms, leveraging their abilities to utilize global statistics and strong local fitting capabilities. Furthermore, we restructure channels into the batch dimension and then group them into multiple sub-features, allowing spatial semantic features to be evenly distributed among sub-features. Through targeted experiments, the effectiveness of our proposed method is validated.