基于分频和多感受野残差密集的指静脉图像超分辨率重建  

Super resolution reconstruction of digital vein image based on frequencydivision and residual of multi receptive field density

作　　者：李礁[1] 钟乐海 包晓安[2] 张娜[2] 邢伟寅[1] 韩正勇 Li Jiao;Zhong Lehai;Bao Xiao’an;Zhang Na;Xing Weiyin;Han Zhengyong(School of Electronics&Information,Mianyang Polytechnic,Mianyang Sichuan 621000,China;School of Information,Zhejiang Sci-Tech University,Hangzhou 310018,China)

机构地区：[1]绵阳职业技术学院电子与信息学院,四川绵阳621000 [2]浙江理工大学信息学院,杭州310018

出　　处：《计算机应用研究》2022年第6期1897-1900,1910,共5页Application Research of Computers

基　　金：四川省科技计划重点研发项目(2019YFG0112);浙江省科技计划重大科技专项重点工业项目(2014C01047);四川省科技计划重点研发项目(2022YFG0206)。

摘　　要：针对红外线CCD摄像头采集指静脉图像较为模糊造成指静脉识别误检率高的问题,提出了基于分频和多感受野残差密集的指静脉图像超分辨率重建方法。该方法构建了图像高低频信息处理子网络,并将RRFDB结构集成到高频子网络中,以RFB为核心的残差密集块设计提升了感受野并降低计算复杂度,更好地保留了原始指静脉图像的线状纹理特征。实验结果表明,该方法能有效改善指静脉图像质量,与SRCNN、VDSR、DRRN等超分辨率重建方法在FV-USM和MMCBNU-6000数据集上进行对比实验,该方法对指静脉特征提取效果好,重建的图像质量高,PSNR与SSIM均优于其他方法。Aiming at the problem that the finger vein image collected by infrared CCD camera is fuzzy,resulting in high error detection rate of finger vein recognition,this paper proposed super resolution reconstruction of digital vein image based on frequency division and residual of multi receptive field density.This method constructed the image high and low frequency information processing sub network,and integrated the RRFDB structure into the high frequency sub network.The residual dense block design with RFB as the core improved the receptive field,reduced the computational complexity,and better retained the linear texture features of the original digital vein image.The experimental results show that this method can effectively improve the image quality of finger vein.Compared with SRCNN,VDSR,DRRN and other super-resolution reconstruction methods on FV-USM and MMCBNU-6000,this method has good effect on finger vein feature extraction,high reconstructed image quality,and PSNR and SSIM are better than other methods.

关 键 词：指静脉识别 频域分离 多感受野残差密集 超分辨率重建 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]