基于深度卷积神经网络的水下偏色图像增强方法

Enhancement Method of Underwater Color Cast ImageBased on Deep Convolutional Neural Network

针对水下图像在采集和传输过程中存在偏色、模糊等问题,提出一种基于深度卷积神经网络的水下偏色图像增强方法,并给出该方法的收敛性分析.首先,在传统U-Net模型的基础上,构建一种基于偏色图像的卷积神经网络模型,不断学习输入图像与输出图像的色彩偏差;其次,通过引用结构相似性的损失函数,使增强后的水下图像与输入的水下图像在内容结构细节上保持高度相似.该方法解决了水下图像偏色、失真的问题.通过对大量的真实水下数据集进行验证,并与其他算法进行对比实验,实验结果表明,用该方法增强后的水下图像不仅在视觉效果上得到了有效提高,同时也保留了原始影像中蕴含的纹理结构,证明该模型在水下图像增强领域实用性较强.

Aiming at the problem of color cast and blur in the process of acquisition and transmission of underwater images,we proposed an enhancement method of underwater color cast image based on deep convolutional neural network,and gave the convergence analysis of the proposed method.Firstly,on the basis of the traditional U-Net model,a convolutional neural network model based on color cast images was constructed to continuously learn the color cast between the input image and the output image.Secondly,by using the loss function of structural similarity,the enhanced underwater image was highly similar to the input underwater image in the details of the content structure.The model solved the problem of color cast and distortion of underwater images.By verifying a large number of real underwater data sets,and compared with other algorithms,the experimental results show that the underwater image enhanced by the method not only improves the visual effect effectively,but also retains the texture structure contained in the original image,which proves that the model has high practicability in the field of underwater image enhancement.