基于层次化字典学习的图像超分辨率方法
|
摘要
由于基于图像块的超分辨方法容易在图像块堆叠为图像的过程中引入模糊,提出基于层次化字典学习的图像超分辨率方法。把经过传统单次字典学习超分辨后的图像作为新的低分辨率图像,并和原始的高分辨图像组成新的高低分辨率训练集进行联合字典训练,得到第二级高低分辨率字典,对低分辨图像实现逐层超分辨的过程。与同类方法不同,特征提取阶段,采用lanczos3插值代替常用的双三次插值方法,利用改进的四方向Sobel算子提取每一小块的特征。与传统的超分辨率方法相比,该方法对低分辨率图像有着较好的细节恢复能力。
Considering that patch-based image super-resolution methods often bring fuzzy effect from patch to image,a hierarchical dictionary learning method for image super-resolution was proposed,where the reconstructed image based on the first-stage dictionaries was treated as the low-resolution input of the second-stage.Different from methods of the same class,in the feature extraction phase,the lanczos3 function was used to interpolate the low-resolution image instead of bicubic interpolation,and the improved four-direction Sobel operator was used to extract features of each patch.Compared with the traditional super-resolution methods,the proposed method has good detail restoration ability for low-resolution images.
Considering that patch-based image super-resolution methods often bring fuzzy effect from patch to image,a hierarchical dictionary learning method for image super-resolution was proposed,where the reconstructed image based on the first-stage dictionaries was treated as the low-resolution input of the second-stage.Different from methods of the same class,in the feature extraction phase,the lanczos3 function was used to interpolate the low-resolution image instead of bicubic interpolation,and the improved four-direction Sobel operator was used to extract features of each patch.Compared with the traditional super-resolution methods,the proposed method has good detail restoration ability for low-resolution images.
引文
[1]CHEN Xiangji.Research on image super-resolution reconstruction algorithm based on multi-scale similarity learning[D].Guangzhou:Institutes of Technology of South China,2015(in Chinese).[陈湘骥.基于多尺度相似学习的图像超分辨率重建算法研究[D].广州:华南理工大学,2015.]
[2]XI Huiqin.Super resolution images reconstruction technique[D].Beijing:Beijing University of Technology,2013(in Chinese).[郗慧琴.超分辨率图像重建技术的研究[D].北京:北京工业大学,2013.]
[3]ZHANG Kaibing,XIA Hongxing,LI Jiping,et al.Structure self similar linear regression superresolution reconstruction[J].Journal of Hubei Engineering University,2014,34(6):42-44(in Chinese).[张凯兵,夏洪星,李纪平,等.结构自相似线性回归超分辨重建[J].湖北工程学院学报,2014,34(6):42-44.]
[4]Hardie R C,Barnard K J.Fast super-resolution using an adaptive Wiener filter with robustness to local motion[J].Optics Express,2012,20(19):21053-21073.
[5]LIU Chao.Research and implementation of fast super-resolution reconstruction method[D].Hangzhou:Electronic University of Science&Technology of Hangzhou,2014(in Chinese).[刘超.快速超分辨率重建方法研究与实现[D].杭州:杭州电子科技大学,2014.]
[6]XU Guoming,XUE Mogen,YUAN Guanglin.Image superresolution reconstruction method via mixture gaussian sparse coding[J].Proceedings of China Optical Engineering,2013,40(3):94-101(in Chinese).[徐国明,薛模根,袁广林.基于混合高斯稀疏编码的图像超分辨率重建方法[J].光电工程,2013,40(3):94-101.]
[7]YU Wei,YAO Hongxun,SUN Xiaoshuai,et al.Contextual dictionary learning for super resolution[J].Computer Science,2014,41(10):87-90(in Chinese).[于伟,姚鸿勋,孙晓帅,等.面向图像超分辨率的上下文字典学习[J].计算机科学,2014,41(10):87-90.]
[8]SONG Jingqi,LIU Hui,ZHANG Caiming.Medical image super-resolution reconstruction based on adaptive patch clustering[J].Computer Science,2016,43(s2):210-214(in Chinese).[宋景琦,刘慧,张彩明.基于自适应块聚类的医学图像超分辨重建[J].计算机科学,2016,43(s2):210-214.]
[9]CHENG Xu.Research on fast 3Dreconstruction algorithm of cone beam CT based on wavelet transform[D].Jinan:Shandong University,2017(in Chinese).[成旭.基于小波变换的锥束CT快速三维重建算法研究[D].济南:山东大学,2017.]
[10]SHEN Dehai,HOU Jian,E Xu,et al.Sobel based edge detection algorithm for multi-direction template[J].Modern Electronics Technique,2015,38(4):91-93(in Chinese).[沈德海,侯建,鄂旭,等.基于Sobel的多方向算子模板边缘检测算法[J].现代电子技术,2015,38(4):91-93.]
[11]YANG Min,LI Min,YI Yaxing.A sparse representation based method for infrared image super-resolution reconstruction[J].Electro-Optical and Control,2016,23(12):1-4(in Chinese).[杨敏,李敏,易亚星.一种改进的稀疏表示红外图像超分辨率重建[J].电光与控制,2016,23(12):1-4.]
[2]XI Huiqin.Super resolution images reconstruction technique[D].Beijing:Beijing University of Technology,2013(in Chinese).[郗慧琴.超分辨率图像重建技术的研究[D].北京:北京工业大学,2013.]
[3]ZHANG Kaibing,XIA Hongxing,LI Jiping,et al.Structure self similar linear regression superresolution reconstruction[J].Journal of Hubei Engineering University,2014,34(6):42-44(in Chinese).[张凯兵,夏洪星,李纪平,等.结构自相似线性回归超分辨重建[J].湖北工程学院学报,2014,34(6):42-44.]
[4]Hardie R C,Barnard K J.Fast super-resolution using an adaptive Wiener filter with robustness to local motion[J].Optics Express,2012,20(19):21053-21073.
[5]LIU Chao.Research and implementation of fast super-resolution reconstruction method[D].Hangzhou:Electronic University of Science&Technology of Hangzhou,2014(in Chinese).[刘超.快速超分辨率重建方法研究与实现[D].杭州:杭州电子科技大学,2014.]
[6]XU Guoming,XUE Mogen,YUAN Guanglin.Image superresolution reconstruction method via mixture gaussian sparse coding[J].Proceedings of China Optical Engineering,2013,40(3):94-101(in Chinese).[徐国明,薛模根,袁广林.基于混合高斯稀疏编码的图像超分辨率重建方法[J].光电工程,2013,40(3):94-101.]
[7]YU Wei,YAO Hongxun,SUN Xiaoshuai,et al.Contextual dictionary learning for super resolution[J].Computer Science,2014,41(10):87-90(in Chinese).[于伟,姚鸿勋,孙晓帅,等.面向图像超分辨率的上下文字典学习[J].计算机科学,2014,41(10):87-90.]
[8]SONG Jingqi,LIU Hui,ZHANG Caiming.Medical image super-resolution reconstruction based on adaptive patch clustering[J].Computer Science,2016,43(s2):210-214(in Chinese).[宋景琦,刘慧,张彩明.基于自适应块聚类的医学图像超分辨重建[J].计算机科学,2016,43(s2):210-214.]
[9]CHENG Xu.Research on fast 3Dreconstruction algorithm of cone beam CT based on wavelet transform[D].Jinan:Shandong University,2017(in Chinese).[成旭.基于小波变换的锥束CT快速三维重建算法研究[D].济南:山东大学,2017.]
[10]SHEN Dehai,HOU Jian,E Xu,et al.Sobel based edge detection algorithm for multi-direction template[J].Modern Electronics Technique,2015,38(4):91-93(in Chinese).[沈德海,侯建,鄂旭,等.基于Sobel的多方向算子模板边缘检测算法[J].现代电子技术,2015,38(4):91-93.]
[11]YANG Min,LI Min,YI Yaxing.A sparse representation based method for infrared image super-resolution reconstruction[J].Electro-Optical and Control,2016,23(12):1-4(in Chinese).[杨敏,李敏,易亚星.一种改进的稀疏表示红外图像超分辨率重建[J].电光与控制,2016,23(12):1-4.]