道路交通模糊图像多尺度清晰化复原方法研究  

Research on multi⁃scale sharpening restoration method of road traffic blurred images

作　　者：吴兰[1] 范晋卿 文成林 WU Lan;FAN Jinqing;WEN Chenglin(College of Electrical Engineering,Henan University of Technology,Zhengzhou 450001,China;School of Automation,Guangdong University of Petrochemical Technology,Maoming 525000,China)

机构地区：[1]河南工业大学电气工程学院,河南郑州450001 [2]广东石油化工学院自动化学院,广东茂名525000

出　　处：《现代电子技术》2022年第9期21-28,共8页Modern Electronics Technique

基　　金：国家自然科学基金资助项目(61973103);河南大学重点科研资助项目(19A120002)。

摘　　要：针对智慧交通监控中由于车速过快或成像距离较远导致的道路交通图像模糊问题,采用生成对抗的思想,提出一种多尺度金字塔盲去模糊网络模型方法。该方法基于特征金字塔原理对道路交通图像进行多尺度和多特征的提取和融合;在局部和全局判别器的基础上提出下采样多尺度判别器,在不同分辨率和随机补丁的图像上进行充分判断;引入多尺度结构相似性损失函数进一步约束高质量图像的生成,并在GoPro和收集整理的道路交通数据集上进行仿真实验验证。仿真结果表明,相较于DeblurGAN和SRN的经典去模糊网络模型,PSNR值最高提升了3.27,SSIM值最高提升了0.12,MOS值最高提升了0.3,在大幅度增强道路交通图像视觉效果的同时,还能够实现监控模糊图像稳定且高质量的复原,在不同道路交通场景下均具有较好的泛化性能。In view of the road traffic image bluring caused by excessively fast vehicle speed or far imaging distance in intelligent traffic monitoring,a multi⁃scale pyramid blind deblurring network model is proposed based on the idea of generative adversarial networks(GANs).On the basis of the principle of feature pyramid,the multi⁃scale and multi⁃feature extraction and fusion of the road traffic images is implemented in this method.A down⁃sampling multi⁃scale discriminator is proposed on the basis of the local and global discriminators to make full judgments on the images with different resolutions and random patches.A multi⁃scale structural similarity loss function is introduced to further restrict the generation of high⁃quality images.The simulation experiments were performed on GoPro dataset and the collected road traffic dataset.The simulation results show that,in comparison with the classic deblurring network models of DeblurGAN and SRN,the PSNR(peak signal⁃to⁃noise ratio)value of the proposed model can be increased by 3.27,its SSIM(structural similarity index)value can be increased by 0.12 and its MOS(mean opinion score)value can be increased by 0.3.The proposed model can realize stable and high⁃quality restoration of monitoring blurred images while greatly enhancing the visual effect of road traffic images.In addition,it has good generalization performance in different road traffic scenes.

关 键 词：模糊图像复原 道路交通 多特征提取 多特征融合 生成对抗网络 深度学习 对比验证 

分 类 号：TN911.73-34[电子电信—通信与信息系统]