Dual-Input Dual-Output Underwater Image Enhancement with Branch Interaction

Due to the inevitable light absorption and scattering, the quality of underwater images is severely compromised, limiting human exploration of the ocean world. To overcome this challenge, we have developed a Dual-branch interactive dual-input dual-output underwater image enhancement model, termed DBINet, which aims to improve the clarity of underwater images and thus provide strong support for underwater activities. The core of DBINet lies in its dual-branch structure, which can simultaneously process inputs from different underwater images and extract common degradation features from these images. Initially, information extraction is performed using an efficient Transformer block. Then, through the Dual-Branch Skip Layer, we achieve interaction and supplementation of features, further enhancing information in similar regions and thus improving the final image quality. Experimental results confirm the effectiveness of DBINet, significantly enhancing the clarity of underwater images and producing higher-quality enhanced images.