Single Image Super-resolution Based on Reversible Network
-
摘要:
本文提出了一种基于可逆神经网络的单一图像超分辨率算法.最近提出的基于深度神经网络的单一图像超分辨率模型都是利用生成的超分辨率图像和对应的高分辨率图像之间的差异定义目标函数以及更新模型参数,这些模型仅利用了超分辨率正向过程中高分辨率图像对于低分辨率图像的依赖,并没有建立低分辨率图像与高分辨率图像之间的相互依赖.本文提出的超分辨可逆网络利用具有可逆结构的神经网络建立低分辨率图像与高分辨率图像之间的相互依赖,它能够将低分辨率图像和高分辨率图像分别投影到相互的图像空间之中,然后利用两个投影的误差反馈来优化模型在低分辨率和高分辨率图像空间的相互映射,基于模型的可逆特性,实现从正向和逆向两个过程分别对超分辨率过程的优化.通过实验,提出的模型在超分辨率基准数据集上取得了优异的结果.
Abstract:A single image super-resolution algorithm based on reversible network is proposed in this paper. Many models based on deep neural network (DNN) are proposed recently to solve the problem of single image super-resolution. In these models, the difference between the generated super-resolution image and the corresponding high-resolution image used to define the objective function and update the model parameters. These methods only take advantage of the dependence of high resolution image on low resolution image, and do not establish the inter-dependence between them. In this paper, we propose a super-resolution reversible network (SRRevnet) model based on reversible network to establish the Mutual mapping between the high-resolution image and low-resolution image. This model maps low resolution image and high resolution image to each other's resolution space respectively, and then use the error feedback to optimize those two opposite process. because the model is reversible, this model can optimize the process of super-resolution from forward and backward respectively. To our knowledge, this paper is the first to use the neural network with reversible structure to solve the problem of single image super- resolution. Through experiments, our model has achieved excellent results on the super-resolution benchmark datasets.
-
Keywords:
- deep learning /
- image process /
- super-resolution /
- reversible network
-
-
表 1 在Set5数据集上的定量对比
Table 1 The quantitative comparison on Set5
set5 nearest bicubic glasner ScSR SRCNN Kim SelfExSR VPGF SRRevnet(ours) PSNR 26.489 7 28.579 9 28.926 9 29.195 8 30.123 0 30.123 8 30.374 0 30.510 0 30.791 8 SSIM 0.764 8 0.821 8 0.833 3 0.839 3 0.862 6 0.866 1 0.873 2 0.863 2 0.883 2 表 2 在Set14数据集上的定量对比
Table 2 The quantitative comparison on Set14
Set14 nearest bicubic glasner ScSR SRCNN Kim SelfExSR VPGF SRRevnet(ours) PSNR 24.493 3 25.789 5 26.199 4 25.763 7 26.354 9 25.239 1 26.768 4 27.23 28.13 SSIM 0.679 7 0.719 5 0.734 9 0.740 5 0.756 7 0.761 0 0.770 46 0.748 6 0.789 5 表 3 在BSD100数据集上的定量对比
Table 3 The quantitative comparison on BSD100
BSD100 nearest bicubic glasner ScSR SRCNN Kim SelfExSR VPGF SRRevnet(ours) PSNR 25.092 6 25.991 8 26.191 9 26.621 5 26.706 5 26.725 7 26.878 1 27.01 27.690 2 SSIM 0.653 4 0.683 7 0.692 2 0.716 0 0.719 6 0.721 4 0.729 8 0.710 5 0.737 9 -
[1] Irani M, Peleg S. Improving resolution by image registration[J]. CVGIP: Graphical Models and Image Processing, 1991, 53(3): 231-239. doi: 10.1016/1049-9652(91)90045-L
[2] Stark H, Oskoui P. High-resolution image recovery from image plane arrays, using convex projections[J]. Journal of the Optical Society of America A Optics & Image Science, 1989, 6(11): 1715-1726. https://pubmed.ncbi.nlm.nih.gov/2585170/
[3] Chang H, Yeung D Y, Xiong Y. Super-resolution through neighbor embedding[C]//IEEE Computer Society Conference on Computer Vision & Pattern Recognition. Piscataway, USA: IEEE, 2004: 275-282.
[4] Gao X B, Zhang K B, Tao D C, et al. Image super-resolution with sparse neighbor embedding[J]. IEEE Transactions on Image Processing, 2012, 21(7): 3194-3205. doi: 10.1109/TIP.2012.2190080
[5] 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: 10.1023/A:1026501619075
[6] Polatkan G, Zhou M, Carin L, et al. A Bayesian nonparametric approach to image super-resolution[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2014, 37(2): 346-358. http://med.wanfangdata.com.cn/Paper/Detail/PeriodicalPaper_PM26353246
[7] Timofte R, De Smet V, Van Gool L. A+: Adjusted anchored neighborhood regression for fast super-resolution[C]//Asian Conference on Computer Vision. Berlin, Germany: Springer, 2014: 111-126.
[8] Hu Y, Wang N, Tao D, et al. SERF: A simple, effective, robust, and fast image super-resolver from cascaded linear regression[J]. IEEE Transactions on Image Processing, 2016, 25(9): 4091-4102. doi: 10.1109/TIP.2016.2580942
[9] Wang H J, Gao X B, Zhang K B, et al. Single-image super-resolution using active-sampling Gaussian process regression[J]. IEEE Transactions on Image Processing, 2015, 25(2): 935-948. https://ieeexplore.ieee.org/document/7364246/citations
[10] Yang J C, Wright J, Huang T S, et al. Image super-resolution via sparse representation[J]. IEEE Transactions on Image Processing, 2010, 19(11): 2861-2873. doi: 10.1109/TIP.2010.2050625
[11] He L, Qi H, Zaretzki R. Beta process joint dictionary learning for coupled feature spaces with application to single image super-resolution[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2013: 345-352.
[12] Schulter S, Leistner C, Bischof H. Fast and accurate image upscaling with super-resolution forests[C]//IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2015: 3791-3799.
[13] Wang Z, Bovik A C, Sheikh H. et al. Image quality assessment: From error visibility to structural similarity[J]. IEEE Transactions on Image Processing, 2004, 13(4): 600-612. doi: 10.1109/TIP.2003.819861
[14] Dong C, Chen C L, He K, et al. Image super-resolution using deep convolutional networks[J]. IEEE Transactions on Pattern Analysis & Machine Intelligence, 2016, 38(2): 295-307. http://ieeexplore.ieee.org/document/7115171
[15] Dong C, Chen C L, Tang X. Accelerating the super-resolution convolutional neural network[C]//European Conference on Computer Vision. Berlin, Germnay: Springer, 2016: 391-407.
[16] Shi W, Caballero J, Huszar F, et al. Real-time single image and video super-resolution using an efficient sub-pixel convolutional neural network[C]//IEEE Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2016: 1874-1883.
[17] Kim J, Lee J K, Lee K M. Accurate image super-resolution using very deep convolutional networks[C]//IEEE Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2016: 1646-1654.
[18] Kim J, Lee J K, Lee K M. Deeply-recursive convolutional network for image super-resolution[C]//IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2015: 1637-1645.
[19] Ledig C, Wang Z, Shi W, et al. Photo-realistic single image super-resolution using a generative adversarial network[C]//IEEE Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2017: 105-114.
[20] Maclaurin D, Duvenaud D, Adams R. Gradient-based hyperparameter optimization through reversible learning[C]//International Conference on Machine Learning. Berlin, Germany: Springer, 2015: 2113-2122.
[21] Dinh L, Krueger D, Bengio Y. Nice: Non-linear independent components estimation[EB/OL]. (2015-04-10)[2019-12-15]. https://arxiv.org/abs/1410.8516.
[22] Dinh L, Sohl-Dickstein J, Bengio S. Density estimation using real-NVP[EB/OL]. (2017-02-27)[2019-12-20]. https://arxiv/abs/1605.08803.
[23] Kingma D P, Dhariwal P. Glow: Generative flow with invertible 1×1 convolutions[C]//Advances in Neural Information Processing Systems. Red Hook, USA: Curran Associates, Inc., 2018: 10215-10224.
[24] Ioffe S, Szegedy C. Batch normalization: Accelerating deep network training by reducing internal covariate shift[EB/OL]. (2015-03-02)[2019-11-29]. https://arxiv/abs/1502.03167.
[25] Grover A, Dhar M, Ermon S. Flow-GAN: Combining maximum likelihood and adversarial learning in generative models[EB/OL]. (2018-01-03)[2019-06-20]. http://arxiv.org/abs/1705.88068.
[26] Goodfellow I J, Pouget-Abadie J, Mirza M, et al. Generative adversarial nets[C]//International Conference on Neural Information Processing Systems. Berlin, Germany: Springer, 2014: 2672-2680.
[27] Bevilacqua M, Roumy A, Guillemot C, et al. Low-complexity single-image super-resolution based on nonnegative neighbor embedding[C]//British Machine Vision Conference. London, UK: Cambridge, 2012: 135.1-135.10.
[28] Zeyde R, Elad M, Protter M. On single image scale-up using sparse-representations[C]//International Conference on Curves and Surfaces, Berlin, Germany: Springer, 2010: 711-730.
[29] Martin D, Fowlkes C, Tal D, et al. A database of human segmented natural images and its application to evaluating segmentation algorithms and measuring ecological statistics[C]//IEEE International Conference on Computer Vision. Piscataway, USA: IEEE, 2001: 416-423.
[30] Kim K I, Kwon Y. Single-image super-resolution using sparse regression and natural image prior[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(6): 1127-1133. doi: 10.1109/TPAMI.2010.25
[31] Glasner D, Bagon S, Irani M. Super-resolution from a single image[C]//IEEE 12th International Conference on Computer Vision. Piscataway, USA: IEEE, 2009: 349-356.
[32] Huang J B, Singh A, Ahuja N. Single image super-resolution from transformed self-exemplars[C]//IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, USA: IEEE, 2015: 5197-5206.
[33] Wang Z, Chen B, Zhang H, et al. Variational probabilistic generative framework for single image super-resolution[J]. Signal Processing, 2019, 156: 92-105. doi: 10.1016/j.sigpro.2018.10.004