-
摘要:为了解决基于字典学习的超分辨重构算法耗时过长的问题，提出了基于稀疏阈值模型的图像超分辨率重建方法。首先，将联合字典理论与图像块稀疏阈值方法相结合，训练得到高、低分辨率过完备图像字典对。接着，通过稀疏阈值OMP算法对图像特征块进行稀疏表示。然后，通过高分辨率字典重构出初始的超分辨图像。最后，通过改进迭代反投影算法对初始的超分辨图像进行全局优化，从而进一步提高图像重构质量。实验结果表明，超分辨图像重构平均峰值信噪比（PSNR）为30.1dB，平均结构自相似度（SSIM）为0.9379，平均计算时间为10.2s。有效提高了超分辨重构的速度，改善了重构高分辨图像的质量。Abstract:In order to solve the problem of the time consuming of the super-resolution reconstruction algorithm based on dictionary learning, a method of super-resolution image reconstruction based on sparse threshold model is proposed. First of all, the over-complete dictionary couple based on the theory of joint dictionary by method of sparse threshold is obtained. And then, the sparse representation of feature block image is represented by sparse threshold OMP algorithm. Then, the initial super-resolution image is reconstructed by the high resolution dictionary. Finally, the global optimization of the initial super-resolution image is improved by the modified iterative back projection algorithm, which can improve the quality of reconstructed image. The experimental results show that the average peak signal to noise ratio(PSNR) is 30.1dB; the average structure self-similarity(SSIM) is 0.9379; the average computation time is 10.2s. This method can improve not only the speed of super-resolution reconstruction, but also the quality of reconstructed high resolution images.
-
图 13倍超分辨率Lena图像重建结果
Figure 1.Results of 3×super-resolultion reconstruction for image of Lena
图 23倍超分辨Pepper局部重建结果
Figure 2.Local details of 3×super resolultion reconstruction for iamge of Pepper
表 13种算法重构图像峰值信噪比(PSNR/dB)和结构自相似度(SSIM)对比
Table 1.Comparison of PSNRs and SSIMs by three methods
标准图像 Bicubic算法 Yang算法 本文算法 Barbara 26.2/0.880 1 26.4/0.887 0 26.7/0.901 5 Bridge 24.4/0.870 5 24.8/0.899 9 24.9/0.903 0 Foreman 31.2/0.908 7 32.0/0.915 3 33.4/0.932 7 Lena 31.7/0.954 9 32.6/0.957 7 32.9/0.968 2 Pepper 32.4/0.969 9 33.3/0.965 1 34.3/0.979 1 Zebra 26.6/0.914 9 28.0/0.935 8 28.6/0.943 0 
下载:
导出CSV
-
[1] KANG M, CHAUDHURIS.Super-resolution image reconstruction[J].IEEE, 2003, 20(3):1920-1935.. [2] 彭真明, 景亮, 何艳敏, 等.基于多尺度稀疏字典的多聚焦图像超分辨融合[J].光学精密工程, 2014, 22(1):169-176.doi:10.3788/OPE.PENG ZH M, JING L, HE Y M, et al..Superresolution fusion of multi-focus image based on multiscale sparse dictionary[J].Opt.Precision Eng., 2014, 22(1):169-176.(in Chinese)doi:10.3788/OPE. [3] FREEMAN W T, PASZTOR E C, CARMICHAEL O T.Learning low-level vision, " IJCV[J].J.Computer Vision, 2000, 40:25-47.doi:10.1023/A:1026501619075 [4] YANG J, WRIGHT J, HUANG T, et al..Image super-resolution via sparse representation[J].IEEE, 2010, 19(11):2861-2873.http://www.idm.pku.edu.cn/staff/zhangjian/Papers/ISCAS2012.pdf [5] 陈健, 高慧斌, 王伟国, 等.超分辨率复原方法相关原理研究[J].中国光学, 2014, 7(6):897-910.//www.illord.com/CN/abstract/abstract9229.shtmlCHEN J, GAO H B, WANG W G, et al..Correlation theory of super-resolution restoration method[J].Chinese Optics, 2014, 7(6):897-910.//www.illord.com/CN/abstract/abstract9229.shtml [6] 张振东, 陈健, 王伟国, 等.基于SSIM_NCCDFT的超分辨率复原评价方法研究[J].液晶与显示, 2015, 30(4):713-721.doi:10.3788/YJYXSZHANG ZH D, CHEN J, WANG W G, et al..Evaluation method of super-resolution restoration based on SSIM_NCCDFT[J].Chinese J.Liquid Crystals and Displays, 2015, 30(4):713-721.(in Chinese)doi:10.3788/YJYXS [7] FREEMAN W T, JONES T R, PASZTOR E C.Example-based super-resolution[J].Computer Graphics & Applications IEEE, 2002, 22(2):56-65.http://academic.research.microsoft.com/Paper/724206 [8] SUN J, ZHENG N, TAO H, et al..Image hallucination with primal sketch priors[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition.IEEE, In Madison, USA, June 18-20, 2003:729-736. [9] 邓承志, 田伟, 汪胜前, 等.近似稀疏正则化的红外图像超分辨率重建[J].光学精密工程, 2014, 22(6):1648-1654.doi:10.3788/OPE.DENG CH ZH, TIAN W, WANG SH Q, et al..Super-resolution reconstruction of approximate sparsity regularized infrared images[J].Opt.Precision Eng., 2014, 22(6):1648-1654.(in Chinese)doi:10.3788/OPE. [10] TANG Y, YUAN Y, YAN P K, et al..Greedy regression in sparse coding sparse for single-image super-resolution[J].J.Vis.Commum.Image R, 2013, 24(2):148-159.doi:10.1016/j.jvcir.2012.02.003 [11] 邓建青, 刘晶红, 刘铁军.基于DSP系统的超分辨率图像重建技术研究[J].液晶与显示, 2012, 27(1):114-120.doi:10.3788/YJYXSDENG J Q, LIU J H, LIU T J.Super-resolution image reconstruction technology based on DSP system[J].Chinese J.Liquid Crystals and Displays, 2012, 27(1):114-120.(in Chinese)doi:10.3788/YJYXS [12] HE X, NIYOGI P.Locality preserving projections[J].Advances in Neural Information Processing System, 2004, 45(1):186-197.http://www.doc88.com/p-0778392992579.html [13] DONOHO D L.Compressed sensing[J].Information Theory IEEE Transactions on, 2006, 52(4):1289-1306.doi:10.1109/TIT.2006.871582 [14] CANDES E, ROMBERG J.Sparsity and incoherence in compressive sampling[J].Inverse Problems, 2006, 23(3):969-985.http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.129.2790 [15] LEE H, BATTLE A, RAINA R, et al..Efficient sparse coding algorithms[J].Nips, 2007:721-728.http://www.doc88.com/p-6909857332289.html [16] CHANG H, YEUNG D Y, XIONG Y M.Super-resolution through neighbor embedding[J].IEEE Conference on Computer Vision & Patter Recognition, 2004, 1:275-282http://cat.inist.fr/?aModele=afficheN&cpsidt=17623200 [17] WANG J, ZHU S, GONG Y.Resolution enhancement based on learning the sparse association of image patches[J].Pattern Recognition Letters, 2010, 31(1):1-10.doi:10.1016/j.patrec.2009.09.004 [18] WANG S, HUANG T Z, LIU J, et al..An alternating iterative algorithm for image deblurring and denoisingproblems[J].Communications in Nonlinear Science & Numerical Simulation, 2014, 19(3):617-626.http://adsabs.harvard.edu/abs/2014CNSNS..19..617W [19] BIOUCAS-DIAS J M, FIGUEIREDO M A T.A new twist:two-step iterative shrinkage/thresholding algorithms for image restoration[J].IEEE Transactions on Image Processing, 2007, 16(12):2992-3004.doi:10.1109/TIP.2007.909319 [20] BRUNET D, VRSCAY E R, WANG Z.On the mathematical properties of the structural similarity index[J].IEEE Transaction on Image Processing, 2012, 21(4):1488-1499.doi:10.1109/TIP.2011.2173206 -
下载:
点击查看大图
图(2)/
表(1)
计量
- 文章访问数:1898
- HTML全文浏览量:433
- PDF下载量:865
- 被引次数:0
