Underwater Image Enhancement Network Based on Multi-channel Hybrid Attention Mechanism

The absorption or scattering of light under water causes problems such as color cast, blur and occlusion in underwater image imaging, which affects underwater vision tasks. Traditional image enhancement methods use histogram equalization, gamma correction and white balance methods to enhance underwater images well. However, there are few studies on the complementarity and correlation of the three methods fused to enhance underwater images. Therefore, an underwater image enhancement network based on multi-channel hybrid attention mechanism is proposed. Firstly, a multi-channel feature extraction module is proposed to extract the contrast, brightness and color features of the image by multi-channel feature extraction of histogram equalization branch, gamma correction branch and white balance branch. Then, the three branch features of histogram equalization, gamma correction and white balance are fused to enhance the complementarity of three branch feature fusion. Finally, a hybrid attention learning module is designed to deeply mine the correlation matrix of the three branches in contrast, brightness and color, and skip connections are introduced to enhance the image output. Experimental results on multiple datasets show that the proposed method can effectively recover the color cast, blur occlusion and improve the brightness of underwater images.