Graph-Based Image Enhancement for Robust Passive Visual Monitoring of Aquatic Ecosystems
Underwater images often suffer significant degradation due to light absorption, scattering, and attenuation, which collectively diminish visual quality. In particular, forward scattering results in blurring of object features, thereby impeding the performance of subsequent computer vision tasks. To address these challenges, this study proposes a U-Net-based architecture for underwater image enhancement. The model incorporates reconstruction losses, including perceptual loss, L1 loss, and mean squared error (MSE), to facilitate effective restoration of image content. To further mitigate the effects of forward scattering, a graph-based convolutional block is integrated into the network. The proposed method is evaluated using the UIEB and EUVP datasets to assess its enhancement performance comprehensively.