基于生成对抗网络的运动模糊图像复原  

Motion Deblurring Based on Generative Adversarial Networks

作　　者：桑亮 高爽 尹增山[1,2,4] SANG Liang;GAO Shuang;YIN Zengshan(Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 200050,China;University of Chinese Academy of Sciences,Beijing 100049,China;School of Information Science and Technology,Shanghai Tech University,Shanghai 201210,China;Shanghai Engineering Center for Microsatellites,Shanghai 201203,China)

机构地区：[1]中国科学院上海微系统与信息技术研究所,上海200050 [2]中国科学院大学,北京100049 [3]上海科技大学信息科学与技术学院,上海201210 [4]上海微小卫星工程中心,上海201203

出　　处：《计算机工程与应用》2019年第6期173-177,共5页Computer Engineering and Applications

基　　金：中国科学院先导培育项目(No.ZDBS16ZRJ1)

摘　　要：针对相机成像时相机抖动、物体运动等导致图像产生运动模糊这一十分具有挑战性的问题,提出基于生成对抗网络的深度卷积神经网络来复原模糊图像的解决方案。该方案省略了模糊核估计的过程,采用端对端的方式直接获取复原图像;通过引入生成对抗网络思想的对抗损失和对残差网络进行改进,有效地复原了图像的细节信息。最后通过训练此深度卷积神经网络模型并在相关模糊复原基准数据集上测试,证明了该方案取得了较好的结果。Image motion blur is a very challenging problem caused by camera shaking or object movements.In order to tackle this problem,the paper proposes a deep convolutional neural network based on generative adversarial networks method.The proposed method can restore a clear image in an end-to-end way without estimating blur kernel.By introducing adversarial loss based on generative adversarial networks and modifying the residual network structure,the proposed method can restore image details effectively.Then this paper trains this deep convolutional neural network model on public datasets.Finally,it is proved that the proposed method achieves good results according to the test on blurry image benchmark datasets.

关 键 词：运动模糊 图像复原 生成对抗网络 深度学习 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]