基于条件生成对抗网络的单图像去雨研究  

Raindrop Removal in a Single Image Based on Conditional Generative Adversarial Networks

作　　者：朱敏[1] 方超 齐美彬[2] ZHU Min;FANG Chao;QI Meibin(Hefei University of Technology,School of Electrical Engineering and Automation,Hefei 230009,China;Hefei University of Technology,School of Computer Science and Information,Hefei 230009,China)

机构地区：[1]合肥工业大学电气与自动化工程学院,合肥230009 [2]合肥工业大学计算机与信息学院,合肥230009

出　　处：《电光与控制》2020年第7期77-82,共6页Electronics Optics & Control

基　　金：国家自然科学基金(61771180)。

摘　　要：雨天会大幅度降低图像质量,对图像的后续处理产生阻碍。为了实现含雨图像的雨滴去除,提出一种基于条件生成对抗网络的单图像去雨滴方法。该方法采用条件生成对抗网络(CGAN)基本架构,以雨滴图像作为附加条件信息,并且添加Lipschitz约束条件;采用条件对抗损失、内容损失以及感知损失相结合的方式来训练网络模型,以修复有雨滴的区域,并重建图像。实验结果表明,提出的方法相较于现有算法雨滴去除效果更好,并且在保证雨滴去除效果的基础上避免图像模糊。Rainy weather will greatly worsen the image quality and hinder the subsequent processing of the image.In order to realize raindrop removal in the image a single-image raindrop removal method based on Conditional Generative Adversarial Networks(CGAN)is proposed.This method adopts the basic framework of CGAN uses the raindrop image as additional condition information and adds Lipschitz constraints.The network model is trained by combining condition adversarial loss content loss with perception loss to repair the raindrop area and reconstruct the image.The experimental results show that the proposed method has better raindrop removal effects than the existing algorithms and can avoid image blurring on the basis of ensuring the raindrop removal effect.

关 键 词：雨滴去除 条件生成对抗网络 Lipschitz约束条件 网络模型 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]