一种新型的基于深度学习的单幅图像去雨方法

Novel Single Image Raindrop Removal Algorithm Based on Deep Learning

雨滴严重影响了图像的视觉效果和后续的图像处理应用。目前,基于深度学习的单幅图像去雨方法能够有效挖掘图像的深度特征,其去雨效果优于传统方法;然而,随着网络深度的增加,网络容易出现过拟合的现象,使得去雨效果遇到瓶颈。文中在继承深度学习优点的基础上,学习有雨/无雨图像之间的残差,然后将残差与源图像进行重构,从而获得无雨图像。该方式大幅增加了网络深度,并加快了算法的收敛速度。分别利用通过不同方式获取的雨滴图像对所提方法进行实验验证,并将该方法与当前最新的去雨滴方法作比较,结果表明所提算法的去雨效果更好。

Raindrops seriously affect the visual effect of images and subsequent image processing applications.At pre-sent,the single image raindrop removal method based on deep learning can effectively mining depth features of image,so its effect of removing rain is better than traditional methods.However,with the increasing of network depth,overfitting is easy to occur,resulting in the bottleneck of rain removal effect.This paper proposed a novel single image raindrop removal algorithm based on deep learning.Firstly,on the basis of inheriting the advantages of deep learning,network learns the residuals between rain images and no-rain images.Secondly,the raindrop-removed image is reconstructed from the residual image and the source image.Through these steps,the depth of network is increased and the convergence speed is accelerated.In terms of performance evaluations,a dataset consisting of images in various scenes was used to test the proposed method,and the results were also compared with those of the state-of-the-art raindrop removal methods.The experimental results show the superiority of the proposed algorithm.