Single-image rain removal is a concern in the field of computer vision because rain streaks may reduce image quality. Images captured in rainy days may suffer from non-uniform rain consisting of different densities, shapes, and sizes. In this paper, we propose a novel single-image deraining method called a Recurrent Memory Unit Network (RMUN) to remove rain streaks from individual images. Unlike existing methods, the RMUN is a recurrent network, which can efficiently utilize the results of the current cycle for the next cycle. In addition, the RMUN employs a Residual Memory Unit Block (RMUB) to extract the features, which means that more attention can be paid to the channels of feature map. A Memory Unit block (MUB) is put in the transform path of the network to keep track of rain details. Different levels of features can be passed in the skip connections between the RMUB and MUB. The extensive experiments show that our proposed method performs better than the state-of-the-art methods on synthetic and real-world datasets.

单图像除雨是计算机视觉领域中的一个问题，因为雨条纹可能会降低图像质量。在雨天拍摄的图像可能会受到密度，形状和大小不同的不均匀降雨的影响。在本文中，我们提出了一种新颖的单图像清除方法，称为循环存储单元网络（RMUN），以消除单个图像中的雨水条纹。与现有方法不同，RMUN是循环网络，可以有效地将当前循环的结果用于下一个循环。此外，RMUN还采用了残余存储单元块（RMUB）来提取特征，这意味着可以更加关注特征图的通道。将存储单元块（MUB）放在网络的转换路径中，以跟踪下雨的细节。可以在RMUB和MUB之间的跳过连接中传递不同级别的功能。广泛的实验表明，我们提出的方法在综合和真实数据集上的性能优于最新方法。