摘要
针对车牌图像分辨率低、视觉质量差等问题,提出一种针对模糊车牌图像的超分辨率重建方法。在FSRCNN的基础上进行如下改进:特征提取阶段采用双通道替代单通道,增强对图像有用特征信息的提取;映射部分使用深度可分离卷积替代原有卷积并减少映射层数,提升训练速度;重建部分采用子像素卷积操作替代反卷积层,抑制反卷积层产生的人工冗余信息。实验结果表明,该方法的重建结果与其他方法相比,图像质量在主观和客观方面都有所改善,训练时间也有所减少。
Aiming at the problems of low resolution and poor visual quality of fuzzy license plate images,this paper proposes a super-resolution reconstruction method for fuzzy license plate images.It improved on the basis of FSRCNN.In the feature extraction stage,dual channels were used instead of single channel to enhance the extraction of useful feature information;in the mapping part,the depth separable convolution was used to replace the original convolution,reduce the number of mapping layers,and improve the training speed;in the reconstruction part,the sub-pixel convolution operation was used to replace the deconvolution layer,and the artificial redundant information generated by the deconvolution layer was suppressed.Experimental results show that compared with other methods,the image quality and training time of our method are improved in both subjective and objective aspects.
作者
田煜
贾瑞生
邓梦迪
赵超越
Tian Yu;Jia Ruisheng;Deng Mengdi;Zhao Chaoyue(College of Computer Science and Engineering,Shandong University of Science and Technology,Qingdao 266590,Shandong,China;Shandong Province Key Laboratory of Wisdom Mine Information Technology,Shandong University of Science and Technology,Qingdao 266590,Shandong,China)
出处
《计算机应用与软件》
北大核心
2020年第11期159-164,228,共7页
Computer Applications and Software
基金
山东省自然科学基金项目(ZR2018MEE008)
山东省重点研发计划项目(2017GSF20115)。
关键词
模糊车牌图像
超分辨率重建
卷积神经网络
深度学习
Fuzzy license plate image
Super-resolution reconstruction
Convolutional neural network
Deep learning
