Underwater Image Enhancement Using Encoder-Decoder Scale Attention Network

Capturing underwater images often results in color distortions and blurriness due to the influence of water, leading to color shifts and haze caused by light propagation. To tackle these issues, we introduce the Encoder-Decoder Scale Attention Network (EDSA-Net) in this paper. The EDSA-Net integrates a Scale-Attention Module (SAM) to facilitate channel weight calculation across scales, thus extracting more informative features from the backbone. Additionally, we introduce a novel Cross-Scale Attention Module (CSAM) to address scale inconsistency and selectively weigh feature maps across scales, capturing various features crucial for underwater image restoration. By enabling interactions among different scales, the network extracts essential features from the images and progressively decodes them to restore clear and sharp underwater images. Experimental results demonstrate that our method achieves state-of-the-art (SoTA) performance and surpasses existing approaches in the field of underwater image restoration.