全局特征双注意力融合对抗网络的水下图像增强

Enhancement of Underwater Images Via Global Feature Dual Attention Fusion Generative Adversarial Network

针对水下图像存在雾气分布不均、光照不均等问题,提出了全局特征双注意力融合对抗网络的水下图像增强算法。首先,利用卷积层不断对输入图像进行下采样,代替平均池化来提取输入图像的全局特征;其次,通过构建全局特征双注意力融合模块,以适应多变的水体环境,更有效地增强不同分布程度的水下图像;最后,在训练中加入条件信息作为限制,提升网络的稳定性。实验结果表明,所提算法与其他经典及最新算法相比具有优势,表明其具有良好的图像增强效果。

Aiming at the problems of heterogeneous fog distribution and uneven illumination of underwater imagesan image enhancement method is proposed based on global feature dual attention fusion generative adversarial network.Firstlythe convolutional layers are used instead of average pooling layers to continuously down-sample the input images and extract the global features.Secondlythe global feature dual attention fusion block is constructedwhich is adaptive to the changing water environment and can enhance underwater information with various dissemination degrees more significantly.Finallyconditional information is incorporated as a restriction in training to increase the network's stability.The results of experiments demonstrate that the proposed algorithm outperforms the classic and latest algorithms and has fine functionality.