Underwater image imbalance attenuation compensation based on attention and self-attention mechanism

The scattering and absorption of light through water will lead to underwater images suffering from low contrast and color variations. With the difference belong wavelength, RGB channels obtained non-uniform information. Although many works for underwater image restoration through CNNs, the color distortions caused by imbalance attenuation have not been addressed in previous contributions. In this paper, we demonstrate that employing green and blue channels to support the red channel to extract more depth features is helpful for underwater image recovery tasks. Further, we discard the previous CNN-based model by proposing a new model based on attention and a self-attention mechanism called underwater restoration attention self-attention (URAS). Our pipeline has achieved better performance than other baseline models on the EUVP dataset.