Key points are not available for this paper at this time.
Underwater images are often degraded due to issues like noise, scattering, and color distortion caused by factors such as light absorption. Recent works using deep neural network (DNN) based approaches have shown promising performance in this domain. This work proposes an U-Net architecture-based network for underwater image enhancement. The proposed network consists of specialized components like inception, residual group and receptive field blocks with pixel and channel attention mechanisms. The paper presents detailed ablation studies involved in designing the network. The proposal is evaluated on two standard datasets (UIEB and EUVP) and compared against 8 different baselines. Extensive experiments show the competitive performance of the proposed model.
Zaidi et al. (Wed,) studied this question.