【摘要】 针对水下图像存在的对比度下降、细节丢失、颜色失真、全局色彩偏移等问题，提出了一种双注意力门融合网络的水下图像增强方法．该方法采用加入了空间注意力机制的U型网络处理输入图像来生成去除噪声且突出特征细节的置信度图，并利用加入了通道注意力机制的卷积神经网络优化图像特征得到有效纠正了色彩偏移的特征图；最后将置信度图与特征图融合实现图像增强．在合成数据集和真实水下图像数据集上的实验结果表明：与现有方法相比，该方法取得了更优的水下图像增强效果且具有更好的泛化能力．更多还原

【Abstract】 An underwater image enhancement method that based on double attention mechanism and gate fusion network is proposed to overcome the problems of local detail loss, low image contrast, color cast and global color deviation in underwater image. In this method, the U-structure network with spatial attention mechanism is used to process the input image to generate a confidence map that removed noise and highlighting feature details, and the convolution neural network with channel attention mechanism is used to refine image features to generate a feature map that effectively correct color cast. Finally, the confidence map and the feature map are fused to achieve image enhancement. The experimental results on the synthetic dataset and the real underwater image dataset show that compared with the existing methods, the method proposed in this paper has better underwater image enhancement performance and generalization ability.更多还原