基于残差块网络的图像去雨算法

为了从有雨图像中恢复图像质量,提出了一种基于残差块网络的海面图像去雨算法。该算法将两种类型的残差块网络相结合,用于提取有雨图像的深层次信息。在训练过程中,学习有雨图像与原图之间的残差,使算法操作的图像目标值域缩小,稀疏性增强。在训练数据集方面,我们采用室外雨天图像数据集和两种加雨算法进行模拟的海面雨天图像数据集,以此扩充训练样本。测试集图像我们选择3种不同类型的雨天场景图像,雨水的类型包括雨线和雨滴。实验结果表明了本文的去雨算法能够适用于不同的雨天场景,去雨处理后图像信噪比评价指标均在35以上,结构相似度均在0.97以上,总体上提升了图像的清晰度。

An image derain algorithm based on the residual block network

To recover image quality from images with rain, a rain removal algorithm for sea surface images is proposed, which is based on the residual block network. The algorithm combines two types of residual block networks for extracting deep-level information of images with rain. During the training process, the residuals between the image with rain and the original image are learned. In this way, the target value domain of the image operated by the algorithm is reduced and the sparsity is enhanced. For the training dataset, we use the outdoor rain image dataset and the sea surface rain image dataset simulated by two rain addition algorithms to expand the training samples. For the test images, three different types of rain scene images are selected, and the types of rain include rain lines and raindrops. Experimental results show that the proposed derain algorithm can be applied to different rain scenes. The signal-to-noise ratio evaluation index of the images after derain processing is above 35, and the structural similarity is above 0. 97. The clarity of the images is generally improved.