The paper introduces an attention-based color consistency underwater image enhancement network (ACC-Net) to address the challenges of color deviation, reduced contrast, and distortion in underwater images. The network consists of three main components: the illumination detail network (ID-Net), the balance stretch module, and the prediction learning module. ID-Net generates texture structure and detail information by decoupling the color image into grayscale and color histograms. The color restoration module (CRM) matches color and content feature information to maintain color consistency. The balance stretch module adjusts color distribution using pixel mean and maximum values. The prediction learning module facilitates context feature interaction to enhance the model's effectiveness. Experiments on three real underwater datasets show that ACC-Net produces more natural enhanced images, outperforming state-of-the-art methods. The paper also reviews related works, categorizing them into a priori-based, model-free, and deep learning-based methods, highlighting the limitations of each approach and the advantages of ACC-Net.The paper introduces an attention-based color consistency underwater image enhancement network (ACC-Net) to address the challenges of color deviation, reduced contrast, and distortion in underwater images. The network consists of three main components: the illumination detail network (ID-Net), the balance stretch module, and the prediction learning module. ID-Net generates texture structure and detail information by decoupling the color image into grayscale and color histograms. The color restoration module (CRM) matches color and content feature information to maintain color consistency. The balance stretch module adjusts color distribution using pixel mean and maximum values. The prediction learning module facilitates context feature interaction to enhance the model's effectiveness. Experiments on three real underwater datasets show that ACC-Net produces more natural enhanced images, outperforming state-of-the-art methods. The paper also reviews related works, categorizing them into a priori-based, model-free, and deep learning-based methods, highlighting the limitations of each approach and the advantages of ACC-Net.