摘要
为了进一步提高遥感图像超分辨效果,提高超分辨重建速度。针对以往稀疏超分辨算法中更容易丢失边缘信息和引入噪声的问题,本文改进了特征提取算子,以对称近邻滤波(SNN)代替高斯滤波,重点解决特征空间中的字典学习问题。首先,根据遥感图像退化模型生成训练样本图像,并分别对高、低分辨率遥感图像进行7×7分块,生成字典训练样本。然后,建立连接高、低分辨率图像空间的双参数联合稀疏字典,将字典学习过程中的稀疏系数分解为系数权值和字典原子的乘积,依据字典原子指标训练和更新字典,得到高低分辨率联合字典映射矩阵。最后,进行遥感图像超分辨稀疏重构。实验结果表明:与当前最先进的稀疏表示超分辨算法相比,本文算法得到的超分辨重建遥感图像的主观效果更好,恢复出更多的地物细节信息;客观评价参数峰值信噪比(PSNR)提高约1.7dB,结构相似性(SSIM)提高约0.016。改进的稀疏表示超分辨算法可以有效地提高遥感图像超分辨效果,同时降低重建时间。
To solve the problems of lost details and added noise in the previous sparse representation image super-resolution,an improved feature extraction algorithm was proposed to improve the image Super-Resolution Reconstruction(SRR)effect.The Gaussian filter was replaced by a symmetric nearest neighbor filter to speed up image super-resolution,and the problem of dictionary learning in the feature space was solved.First,sample training images were generated based on the remote sensing image degradation model,and high-low resolution images were divided into image patches sized 7×7.Then,a high-low resolution joint dictionary mapping matrix was generated after the dictionary was trained and updated.Finally,image super-resolution reconstruction was performed in sparse representation.Experimental results revealed that the proposed method reconstructed a higher-quality superresolution image in less time.Simultaneously,as compared with the image obtained with the most advanced sparse representation super-resolution algorithm,the SRR resulting image contained more texture details of ground objects.In addition,the peak signal-to-noise ratio and structural similarity index measure were increased by approximately 1.7 dB and 0.016,respectively.Conclusion:The improved sparse representation SRR algorithm can effectively improve the SRR effect of remote sensing images and reduce the super-resolution reconstruction time.
引文
[1]ZHU F Z,LI J Z,ZHU B,et al..Super-resolution image reconstruction based on three-step-training neural networks[J].Systems Engineering and E-lectronics,2010,21(6):1-7.
[2]朱福珍,李金宗,朱兵,等.基于径向基函数神经网络的超分辨率图像重建[J].光学精密工程,2010,18(6):1444-1451.ZHU F ZH,LI J Z,ZHU B,et al..Super-resolution image reconstruction based on RBF neural network[J].Opt.Precision Eng.,2010,18(6):1444-1451.(in Chinese)
[3]朱福珍,王晓飞,丁群,等.三级训练BP神经网络遥感图像超分辨重建[J].光学精密工程,2015,23(10):208-215.ZHU F ZH,WANG X F,DING Q,et al..Superresolution reconstruction of remote images based on three level training BP neural network[J].Opt.Precision Eng.,2015,23(10):208-215.(in Chinese)
[4]TSAI R Y,HUANG T S.Multiple frame image restoration and registration[J].Advances in Computer Vision and Image Processing Greenwich,1984,1(2):31-35.
[5]LING F,ZHANG Y,FOODY G M,et al..Learning-Based super-resolution land cover mapping[J].IEEE Transactions on Geoscience and Remote Sensing,2016,54(7):3794-3810.
[6]YUAN Y,ZHENG X,LU X.Hyper-spectral image super-resolution by transfer learning[J].IEEEJournal of Selected Topics in Applied Earth Observations and Remote Sensing,2017,10(5):1963-1974.
[7]ZHAO L,SUN Q,ZHANG Z.Single image superresolution based on deep learning features and dictionary model[J].IEEE Access,2017,5(99):17126-17135.
[8]FREEMAN W T,JONES T R,PASZTOR E C.Exampled-based Super-resolution[J].IEEE Computer Society Press,2002,22(2):56-65.
[9]TURKAN M,GUILLEMOT C.Online dictionaries for image prediction[C].IEEE International Conference on Image Processing,2011:293-296.
[10]SCHOLKOPF B,PLATT J,HOFMANN T.Efficient sparse coding algorithms[C].International Conference on Neural Information Processing Systems.MIT Press,2006:801-808.
[11]SUN Y C,GU G H,SUI X B,et al..Single image super-resolution using compressive sensing with a redundant dictionary[J].IEEE Photonics Journal,2015,7(2):1-11.
[12]YANG S Y,SUN F H,WANG M,et al..Novel super resolution restoration of remote sensing images based on compressive sensing and example patches-aided dictionary learning[C].Proceedings of the 2011 International Workshop on multi-Platform and Multi-Sensor Remote Sensing and Mapping.Piscataway,2011:1-6.
[13]陈伟业,孙权森.结合压缩感知与非局部信息的图像超分辨率重建[J].计算机应用,2016,36(9):2570-2575.CHEN W Y,SUN Q S.Image super-resolution reconstruction combined with compressed sensing and nonlocal information[J].Journal of Computer Applications,2016,36(9):2570-2575.(in Chinese)
[14]YEGANLI F,NAZZAL M,UNAL M,et al..Image super-resolution via sparse representation over multiple learned dictionaries based on edge sharpness[J].Signal,Image and Video Processing,2015,10(3):535-542.
[15]WU W,YANG X M,LIU K,et al..A new framework for remote sensing image super-resolution:sparse representation-based method by processing dictionaries with multi-type features[J].Journal of Systems Architecture,2016,6(4):63-75.
[16]穆少硕,张叶,贾平.基于自学习局部线性嵌入的多幅亚像元超分辨成像[J].光学精密工程,2015,23(9):2677-2686.MU SH SH,ZHANG Y,JIA P.Super-resolution imaging of multi-frame sub-pixel images based on self-learning LLE[J].Opt.Precision Eng.,2015,23(9):2677-2686.(in Chinese)
[17]ZEYDE ROMAN,MICHAEL ELAD,MATANPROTTER.On single image scale-up using sparse-representations[C].International Conference on Curves and Surfaces.Springer-Verlag,2010:711-730.
[18]YANG J C,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,2008:1-8.
[19]YANG J C,WANG Z W,LIN Z.Bilevel sparse coding for coupled feature spaces[C].IEEE Conference on Computer Vision and Pattern Recognition,2012:2360-2367.
[20]DONOHO D L,JOHNSTONE I M.Adapting to unknown smoothness via wavelets shrinkage[J].Journal of the American Statistical Association,1995,90(432):1200-1224.
[21]DONOHO D L,JOHNSTONE I M..Ideal spatial adaptation by wavelet shrinkage[J].Biometrika,1994,81(3):425-455.