【摘要】 由于水下小颗粒、浮游生物造成光衰减、散射现象，从而导致水下图像存在雾霾效应及模糊，大大影响人眼对水下图像的视觉感知。为了解决这2个问题，提出一个基于跨色彩空间Transformer的水下图像增强模型(简称为Tcolor)。在这个模型中，采用RGB、HSV、LAB三色彩空间U-Net进行特征提取，通过通道传输注意力模块突出最具有区分度的通道和区域。此外，使用跨色彩空间Transformer模块保证增强图像的全局一致性，有效避免伪影的产生，同时在Transformer模块中加入卷积保留局部细节。最后，提出一个联合损失函数进一步提升模型的能力。在2种公开的数据集上进行对比实验，与次优的数据相比，评价指标PSNR提高了6.87%,SSIM提高了6.22%,UIQM提高了2.53%。在主观感知和客观评估中，增强后的图像在细节、对比度和一致性方面均得到了提升。更多还原

【Abstract】 Due to the presence of small underwater particles and plankton, which cause light attenuation and scattering phenomenon, which leads to underwater image haze effect and blurring, greatly affecting the visual perception of the human eye on the underwater image. In order to solve these two problems, an underwater image enhancement model based on Transformer across color space(referred to as Tcolor) is proposed. In this model, the three color spaces U-Net of RGB, HSV and LAB are used for feature extraction, and the most discriminative channels and regions are highlighted by the Channel Transformer Attention module. In addition, a cross-color space Transformer module is used to ensure the global consistency of the enhanced image and effectively avoid artifacts, while a convolution is added to the Transformer module to preserve local details. Finally, a joint loss function is proposed to further enhance the capability of the model. Comparison experiments are conducted on 2 publicly available datasets, and compared with the suboptimal data, the evaluation metrics PSNR improves by 6.87%, SSIM improves by 6.22%, and UIQM improves by 2.53%.In both subjective perception and objective evaluation, the enhanced images are improved in terms of detail, contrast and consistency.更多还原