采用深度学习的快速超分辨率图像重建方法
|
摘要
为满足实际工业生产需要,提出一种基于深度学习的快速超分辨率图像重建方法.采用一种快速的卷积神经网络结构,使用级联的小卷积核以取得重建速度上的提升,加深卷积网络以取得重建质量上的提升.实验结果表明:在标准的公共数据集上,该算法重建的高分辨率图像在主观视觉感受和客观的图像质量评价(峰值信噪比)上取得较好的效果,且重建时间大大缩短;将算法应用在实际的项目中,能达到阈值分割后准确检测物体的标准,减少企业对高额工业相机的经济开支.
In order to meet the needs of actual industrial production,a fast super-resolution image reconstruction method based on deep learning is proposed.We proposed our own convolutional neural network structure,using cascaded small convolution kernels to achieve a higher reconstruction speed,and deepening the convolution network to achieve an improvement in reconstruction quality.The experimental results show that on the standard public dataset,the high-resolution image reconstructed by the our algorithm achieves better results in subjective visual perception and objective image quality evaluation(peak signal-to-noise ratio),at the mean time,the reconstruction time is greatly shortened.The algorithm is applied in projects to solve the problem for accurately detecting objects after threshold segmentation.In this way,it also reduces the high expenses of enterprises for purchasing industrial cameras.
In order to meet the needs of actual industrial production,a fast super-resolution image reconstruction method based on deep learning is proposed.We proposed our own convolutional neural network structure,using cascaded small convolution kernels to achieve a higher reconstruction speed,and deepening the convolution network to achieve an improvement in reconstruction quality.The experimental results show that on the standard public dataset,the high-resolution image reconstructed by the our algorithm achieves better results in subjective visual perception and objective image quality evaluation(peak signal-to-noise ratio),at the mean time,the reconstruction time is greatly shortened.The algorithm is applied in projects to solve the problem for accurately detecting objects after threshold segmentation.In this way,it also reduces the high expenses of enterprises for purchasing industrial cameras.
引文
[1]曾凯,丁世飞.图像超分辨率重建的研究进展[J].计算机工程与应用,2017,53(16):29-35.DOI:10.3778/j.issn.1002-8331.1705-0097.
[2]YANG C Y,MA Chao,YANG M H.Single-image super-resolution:A benchmark[C]∥European Conference on Computer Vision.Switzerland:Springer,2014:372-386.DOI:10.1007/978-3-319-10593-2_25.
[3]YANG Jianchao,WRIGHT J,HUANG T,et al.Image super-resolution as sparse representation of raw image patches[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Anchorage:IEEE Press,2008:1-8.DOI:10.1109/CVPR.2008.4587647.
[4]HE Hu,KONDI L P.A regularization framework for joint blur estimation and super-resolution of video sequences[C]∥Proceedings of the 2005IEEE International Conference on Image Processing.Genova:IEEE Press,2005:329-332.DOI:10.1109/ICIP.2005.1530395.
[5]LI Xiaoguang,LAM K M,QIU G P,et al.Examplebased image super-resolution with class-speciflc predictors[J].Journal of Visual Communication and Image Representation,2009,20(5):312-322.DOI:10.1016/j.jvcir.2009.03.008.
[6]SU Congyong,ZHUANG Yueting,LI Huang,et al.Steerable pyramid-based face hallucination[J].Pattern Recognition,2005,38(6):813-824.DOI:10.1016/j.patcog.2004.11.007.
[7]JIJI C V,CHAUDHURI S.Single-frame image super-resolution through contourlet learning[C]∥EURASIP Journal on Advances in Signal Processing.New York:Springer,2006:235.DOI:10.1155/ASP/2006/73767.
[8]张晓燕,秦龙龙,钱渊,等.一种改进的稀疏表示超分辨率重建算法[J].重庆邮电大学学报(自然科学版),2016,28(3):400-405.DOI:10.3979/j.issn.1673-825X.2016.03.020.
[9]FREEMAN W T,PASZTOR E C,CARMICHAEL O T.Learning low-level vision[J].International Journal of Computer Vision,2000,40(1):25-47.DOI:10.1109/ICCV.1999.790414.
[10]FREEMAN W T,JONE T R,PASZTOR E C.Example-based super-resolution[J].IEEE Computer Graphics and Applications,2002,22(2):56-65.DOI:10.1109/38.988747.
[11]DONG Chao,CHEN C L,HE Kaiming,et al.Image super-resolution using deep convolutional networks[J].IEEETransactions on Pattern Analysis and Machine Intelligence,2016,38(2):295-307.DOI:10.1109/TPAMI.2015.2439281.
[12]刘晨羽,蒋云飞,李学明.基于卷积神经网的单幅图像超分辨率重建算法[J].计算机辅助设计与图形学学报,2017,29(9):1643-1649.DOI:10.3969/j.issn.1003-9775.2017.09.007.
[13]KIM J,LEE J K,LEE K M.Accurate image super-resolution using very deep convolutional networks[C]∥IEEEConference on Computer Vision and Pattern Recognition.Las Vegas:IEEE Press,2016:1646-1654.DOI:10.1109/CVPR.2016.182.
[14]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[J/OL].Computer Science,2014.[2017-06-10].http:∥arxiv.org/pdf/1409.1556.
[15]SHI Wenzhe,CABALLERO J,HUSZAR F,et al.Real-time single image and video super-resolution using an efficient subpixel convolutional neural network[C]∥IEEE Conference on Computer Vision and Pattern Recognitionhttp.Las Vegas:IEEE Press,2016:1874-1883.DOI:10.1109/CVPR.2016.207.
[16]OTSU N.A threshold selection method from gray-level histograms[J].Automatica,1979,9(1):62-66.DOI:10.1109/TSMC.1979.4310076.
[17]XU Xiaohong,WU Zhihui,CHEN Yu,et al.Plant root spatial distribution measurements based on the hough transformation[J].Neurocomputing,2014,145:209-220.DOI:10.1016/j.neucom.2014.05.041.
[2]YANG C Y,MA Chao,YANG M H.Single-image super-resolution:A benchmark[C]∥European Conference on Computer Vision.Switzerland:Springer,2014:372-386.DOI:10.1007/978-3-319-10593-2_25.
[3]YANG Jianchao,WRIGHT J,HUANG T,et al.Image super-resolution as sparse representation of raw image patches[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Anchorage:IEEE Press,2008:1-8.DOI:10.1109/CVPR.2008.4587647.
[4]HE Hu,KONDI L P.A regularization framework for joint blur estimation and super-resolution of video sequences[C]∥Proceedings of the 2005IEEE International Conference on Image Processing.Genova:IEEE Press,2005:329-332.DOI:10.1109/ICIP.2005.1530395.
[5]LI Xiaoguang,LAM K M,QIU G P,et al.Examplebased image super-resolution with class-speciflc predictors[J].Journal of Visual Communication and Image Representation,2009,20(5):312-322.DOI:10.1016/j.jvcir.2009.03.008.
[6]SU Congyong,ZHUANG Yueting,LI Huang,et al.Steerable pyramid-based face hallucination[J].Pattern Recognition,2005,38(6):813-824.DOI:10.1016/j.patcog.2004.11.007.
[7]JIJI C V,CHAUDHURI S.Single-frame image super-resolution through contourlet learning[C]∥EURASIP Journal on Advances in Signal Processing.New York:Springer,2006:235.DOI:10.1155/ASP/2006/73767.
[8]张晓燕,秦龙龙,钱渊,等.一种改进的稀疏表示超分辨率重建算法[J].重庆邮电大学学报(自然科学版),2016,28(3):400-405.DOI:10.3979/j.issn.1673-825X.2016.03.020.
[9]FREEMAN W T,PASZTOR E C,CARMICHAEL O T.Learning low-level vision[J].International Journal of Computer Vision,2000,40(1):25-47.DOI:10.1109/ICCV.1999.790414.
[10]FREEMAN W T,JONE T R,PASZTOR E C.Example-based super-resolution[J].IEEE Computer Graphics and Applications,2002,22(2):56-65.DOI:10.1109/38.988747.
[11]DONG Chao,CHEN C L,HE Kaiming,et al.Image super-resolution using deep convolutional networks[J].IEEETransactions on Pattern Analysis and Machine Intelligence,2016,38(2):295-307.DOI:10.1109/TPAMI.2015.2439281.
[12]刘晨羽,蒋云飞,李学明.基于卷积神经网的单幅图像超分辨率重建算法[J].计算机辅助设计与图形学学报,2017,29(9):1643-1649.DOI:10.3969/j.issn.1003-9775.2017.09.007.
[13]KIM J,LEE J K,LEE K M.Accurate image super-resolution using very deep convolutional networks[C]∥IEEEConference on Computer Vision and Pattern Recognition.Las Vegas:IEEE Press,2016:1646-1654.DOI:10.1109/CVPR.2016.182.
[14]SIMONYAN K,ZISSERMAN A.Very deep convolutional networks for large-scale image recognition[J/OL].Computer Science,2014.[2017-06-10].http:∥arxiv.org/pdf/1409.1556.
[15]SHI Wenzhe,CABALLERO J,HUSZAR F,et al.Real-time single image and video super-resolution using an efficient subpixel convolutional neural network[C]∥IEEE Conference on Computer Vision and Pattern Recognitionhttp.Las Vegas:IEEE Press,2016:1874-1883.DOI:10.1109/CVPR.2016.207.
[16]OTSU N.A threshold selection method from gray-level histograms[J].Automatica,1979,9(1):62-66.DOI:10.1109/TSMC.1979.4310076.
[17]XU Xiaohong,WU Zhihui,CHEN Yu,et al.Plant root spatial distribution measurements based on the hough transformation[J].Neurocomputing,2014,145:209-220.DOI:10.1016/j.neucom.2014.05.041.