CA-Unet: Cascaded Attention Transformer for Underwater Image Deblurring and Color Restoration

The absorption and scattering of light in water result in low-quality photographs being captured underwater. This degradation hampers analysis in fields such as marine biology. To address this, we propose a novel deep-learning model that utilizes a Cascaded Attention Transformer for underwater image enhancement (CA-Unet). The central component of CA-Unet, the Residual Cascaded Attention Block (RCAB), employs various attention techniques to capture both local and global dependencies while extracting deep and superficial features. This approach effectively recovers lost details in underwater scenes, enhances contrast, reduces color cast, and restores image quality. Compared to other methods, our model demonstrates superior performance on both synthetic and real-world underwater photos, showcasing the effectiveness of the proposed method.