卷积神经网络超分辨率图像重建算法的改进
|
摘要
随着时代与科技的进步,人们对图像分辨率的要求越来越高。提高图像分辨率成为必须要解决的问题。目前利用深度学习进行超分辨率图像重建成为提高图像质量的一种趋势。深度学习对图像进行重建可以有效地提高图像质量。现有的基于卷积神经网络的超分辨率图像重建算法有着自身的优势同时也存在着缺陷。针对算法的缺陷,本文提出一种改进的图像重建算法。系统地分析了卷积神经网络在图像重建时的缺陷,针对重建时的训练时间长,存在网络退化现象等缺点。本文利用残差网络对传统的SRCNN进行改进。改进后的算法与传统的SRCNN算法相比,可以减少训练时间,同时可以防止网络退化现象的发生。
With the advancement of the times and technology, people are increasingly demanding image quality. Increasing image resolution has become a problem that must be solved. At present, the use of deep learning for super-resolution image reconstruction has become a trend to improve image quality. Deep learning to reconstruct images can effectively improve image quality. The existing super-resolution image reconstruction algorithms based on convolutional neural networks have their own advantages and defects. Aiming at the defects of the algorithm,this paper proposes an improved image reconstruction algorithm. The system analyzes the defects of convolutional neural network in image reconstruction, and has shortcomings such as long training time and network degradation phenomenon. This paper uses the residual network to improve the traditional SRCNN. Compared with the traditional SRCNN algorithm, the improved algorithm can reduce the training time and prevent network degradation.
With the advancement of the times and technology, people are increasingly demanding image quality. Increasing image resolution has become a problem that must be solved. At present, the use of deep learning for super-resolution image reconstruction has become a trend to improve image quality. Deep learning to reconstruct images can effectively improve image quality. The existing super-resolution image reconstruction algorithms based on convolutional neural networks have their own advantages and defects. Aiming at the defects of the algorithm,this paper proposes an improved image reconstruction algorithm. The system analyzes the defects of convolutional neural network in image reconstruction, and has shortcomings such as long training time and network degradation phenomenon. This paper uses the residual network to improve the traditional SRCNN. Compared with the traditional SRCNN algorithm, the improved algorithm can reduce the training time and prevent network degradation.
引文
[1] Tian Y,Yap K. Joint image registration and super-resolution from low-resolution images with zooming motion[J]. IEEE Trans on Circuits and Systems for Video Technology,2013,23(7):1224-1234.
[2] Yang Min-Chun, Wang Yu-Chiang Frank. A self-learning approach to single image super-resolution[J]. IEEE Trans on Multimedia,2013,15(3):498-508.
[3] Elad M, Aharon M. Image denoising via sparse and redundant representations over learned dictionaries[J]. IEEE Trans on Image Processing,2006 15(12):3736-3745.
[4] Hou H,Andrews H. Cubic spline for image interpolation and digital filtering[J]. IEEE Transactions Signal Processing,1978,26(6):508-517.
[5] KeysR. Cubic convolution interpolation for digital image processing[J]. IEEE Transactions on Acoustics,Speech and Signal Processing,1981,29(6):1153-1160.
[6] Irani M,Peleg S.Improving resolution by image registration[J].CV-GIP:Graphical Models and Image Processing,1991,53(3):231-239.
[7] Stark H,Oskoui P.High-resolution image recovery form imageplane arrays,using convex projection[J].Journal of the Opt.Soc.of America,1989,6(11):1715-1726.
[8] Schultz R,Stevenson R.Bayesian estimation of sub-pixel motion field and high-resolution video stills[C]//IIEEE Int.conf.on Image Processing,1997(3):62-65.
[9] Fang B, Huang Z, Li Y, Wang Y. v-Support vector machine based on discriminant sparse neighborhood preserving embedding.Pattern Analysis and Applications,2017,20(4):1077-1089.
[10] Yang J,Wang Z,Lin Z,Huang T.Coupled dictionary training for image super-resolution. IEEE Trans. on Image Processing,2012,21(8):3467-3478.
[11] Dong C,Loy CC,He K,Tang X.Image super-resolution using deep convolutional networks. IEEE Trans. on Pattern Analysis and Machine Intelligence,2016,38(2):295-307.
[12] Dong C, Loy CC, Tang X. Accelerating the super-resolution convolutional neural network.In:Proc.of the European Conf.on Computer Vision.Springer-Verlag,2016.391-407.
[13]周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报,2017,40(7):1-23.
[14] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]. IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, 2016:770-778.
[15] ZAGORUYKO S, KOMODAKIS N. Wide residual networks[C]. British Machine Vision Conference. 2016:87.
[2] Yang Min-Chun, Wang Yu-Chiang Frank. A self-learning approach to single image super-resolution[J]. IEEE Trans on Multimedia,2013,15(3):498-508.
[3] Elad M, Aharon M. Image denoising via sparse and redundant representations over learned dictionaries[J]. IEEE Trans on Image Processing,2006 15(12):3736-3745.
[4] Hou H,Andrews H. Cubic spline for image interpolation and digital filtering[J]. IEEE Transactions Signal Processing,1978,26(6):508-517.
[5] KeysR. Cubic convolution interpolation for digital image processing[J]. IEEE Transactions on Acoustics,Speech and Signal Processing,1981,29(6):1153-1160.
[6] Irani M,Peleg S.Improving resolution by image registration[J].CV-GIP:Graphical Models and Image Processing,1991,53(3):231-239.
[7] Stark H,Oskoui P.High-resolution image recovery form imageplane arrays,using convex projection[J].Journal of the Opt.Soc.of America,1989,6(11):1715-1726.
[8] Schultz R,Stevenson R.Bayesian estimation of sub-pixel motion field and high-resolution video stills[C]//IIEEE Int.conf.on Image Processing,1997(3):62-65.
[9] Fang B, Huang Z, Li Y, Wang Y. v-Support vector machine based on discriminant sparse neighborhood preserving embedding.Pattern Analysis and Applications,2017,20(4):1077-1089.
[10] Yang J,Wang Z,Lin Z,Huang T.Coupled dictionary training for image super-resolution. IEEE Trans. on Image Processing,2012,21(8):3467-3478.
[11] Dong C,Loy CC,He K,Tang X.Image super-resolution using deep convolutional networks. IEEE Trans. on Pattern Analysis and Machine Intelligence,2016,38(2):295-307.
[12] Dong C, Loy CC, Tang X. Accelerating the super-resolution convolutional neural network.In:Proc.of the European Conf.on Computer Vision.Springer-Verlag,2016.391-407.
[13]周飞燕,金林鹏,董军.卷积神经网络研究综述[J].计算机学报,2017,40(7):1-23.
[14] HE K, ZHANG X, REN S, et al. Deep residual learning for image recognition[C]. IEEE Conference on Computer Vision and Pattern Recognition. IEEE Computer Society, 2016:770-778.
[15] ZAGORUYKO S, KOMODAKIS N. Wide residual networks[C]. British Machine Vision Conference. 2016:87.