HUANG Zhangyu
（Department of Electronic, Electrical and Systems Engineering, University of Birmingham, Birmingham B152TT, United Kingdom）
Abstract： Single image super-resolution (SISR) is a fundamentally challenging problem because a low-resolution (LR) image can correspond to a set of high-resolution (HR) images, while most are not expected. Recently, SISR can be achieved by a deep learning-based method. By constructing a very deep super-resolution convolutional neural network (VDSRCNN), the LR images can be improved to HR images. This study mainly achieves two objectives: image super-resolution (ISR) and deblurring the image from VDSRCNN. Firstly, by analyzing ISR, we modify different training parameters to test the performance of VDSRCNN. Secondly, we add the motion blurred images to the training set to optimize the performance of VDSRCNN. Finally, we use image quality indexes to evaluate the difference between the images from classical methods and VDSRCNN. The results indicate that the VDSRCNN performs better in generating HR images from LR images using the optimized VDSRCNN in a proper method.
Key words： single image super-resolution (SISR); very deep super-resolution convolutional neural network (VDSRCNN); motion blurred image; image quality index
References
［1］PARK S C, PAR M K, KANG M G. Super-resolution image reconstruction: A technical overview. IEEE Signal Processing Magazine, 2003, 20(3): 21-36.
［2］ZHANG Y F, FAN Q L, BAO F X, et al. Single-image super-resolution based on rational fractal interpolation. IEEE Transactions on Image Processing, 2018, 27(8): 3782-3797.
［3］HE K M, ZHANG X Y, REN S Q, et al. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification//IEEE International Conference On Computer Vision, Dec.7-13, 2018, Santiago, Chile. New York: IEEE, 2018: 1026-1034.
［4］KIM J, LEE J K, LEE K M. Accurate image super-resolution using very deep convolutional networks//IEEE Conference on Computer Vision and Pattern Recognition, Jun.26-Jul.01, 2016, Las Vegas, USA. New York: IEEE, 2016: 1646-1654.
［5］RAMPAL H, DAS A, MAHAJAN S, et al. Enhancement of corn image quality using very-deep super-resolution (VDSR) neural network//The 3rd International Conference on Inventive Computation Technologies, Nov.15-16, 2018, Coimbatore, Tamilnadu, India. Network: IEEE, 2018: 1-4.
［6］HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks for visual recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2015, 37(9): 1904-1916.
［7］DONG C, LOY C C, HE K, et al. Image super-resolution using deep convolutional networks. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2016, 38(2): 295-305.
［8］LEE D, LEE S, LEE H S, et al. Context-preserving filter reorganization for VDSR-based super-resolution//2019 IEEE International Conference on Artificial Intelligence Circuits and Systems, Mar.18-20, 2019, Taiwan, China. New York: IEEE, 2019: 107-111.
［9］KIM J, SHIN M, KIM D, et al. Performance comparison of SRCNN, VDSR, and SRDenseNet deep learning models in embedded autonomous driving platforms//2021 International Conference on Information Networking, Jan.13-16, 2021, Jeju Island, South Korea. Network: IEEE, 2021: 56-58.
［10］DENGWEN Z. An edge-directed bicubic interpolation algorithm//The 3rd International Congress on Image and Signal Processing, Oct.16-18, 2010, Yantai, China. Network: IEEE, 2010:1186-1189.
［11］KREIS, R. Issues of spectral quality in clinical 1H-magnetic resonance spectroscopy and a gallery of artifacts. NMR in Biomedicine, 2004, 17: 361-381.
［12］HORE A, ZIOU D. Image quality metrics: PSNR vs. SSIM//The 20th International Conference on Pattern Recognition, Aug.23-26, 2010, Istanbul, Turkey. New York: IEEE, 2010: 2366-2369.
［13］WANG Z, BOVIK A C, SHEIKH H R, et al. Simoncelli. Image quality assessment: from error visibility to structural similarity. IEEE Transactions on Image Processing, 2004, 13(4): 600-613.
［14］BORA R, SHAHANE N M. Image forgery detection through motion blur estimates//2012 IEEE International Conference on Computational Intelligence and Computing Research, Dec.18-20, 2012. Coimbatore, India. New York: IEEE, 2012: 1-4.
［15］YANG L. Image restoration from a single blurred photograph//The 3rd International Conference on Information Science and Control Engineering, Jul.08-10, 2016, Beijing, China. New York: IEEE, 2016: 405-409.
［16］DONG X, AXINTE D, PALMER D, et al. Development of a slender continuum robotic system for on-wing inspection/repair of gas turbine engines. Robotics and Computer-Inte Grated Manufacturing, 2017, 44: 218-229.
［17］JING Z, QIAO L, PAN H, et al. An overview of the configuration and manipulation of soft robotics for on-orbit servicing. Science China-Information Science, 2017,60: 1-19.
［18］LIU W L, YU D Z, HU J L. Sensitivity analysis of parameters in NASGRO crack growth life model. Journal of Ordnance Equipment Engineering, 2021, 42(11): 77-82.
［19］HUANG R J, RUAN W J , BU P F. Study on opening mechanism of aluminum alloy closure for tube shape seapon. Journal of Ordnance Equipment Engineering, 2021, 42(11): 52-56.
基于极深超分辨率卷积神经网络的单一图像超分辨率研究黄璋豫
（伯明翰大学 电子电气与系统工程系， 英国 B152TT）
摘要:单一图像超分辨率（Sing image super-resolution, SISR）是一个具有挑战性的问题， 从根本上说， 低分辨率图像可以对应于一组高分辨率图像， 但是其困难在于大多数不可预期。 深度学习为单一图像超分辨率研究提供了有效途径， 通过构造极深超分辨卷积神经网络（Very deep super-resolution conventional neural network, VDSRCNN）， 可以将低分辨率图像优化为高分辨率图像。 本研究主要实现了两个目标： 图像超分辨率（Image super-resolution, ISR）和应用极深超分辨率卷积神经网络使图像更清晰。 首先， 在分析ISR的基础上， 对不同的训练参数进行改良， 以测试VDSRCNN的性能； 其次， 通过在训练集中加入运动模糊图像来优化VDSRCNN的结构参数； 最后， 利用图像质量指数评价了传统方法和VDSRCNN方法的图像差异。 结果表明， 本文提出的方法可以有效改良VDSRCNN的结构， 更好地从低分辨率图像生成高分辨率图像。
关键词:单图像超分辨率； 极深超分辨卷积神经网络； 运动模糊图像； 图像质量指数
引用格式：HUANG Zhangyu. Research on single image super-resolution based on very deep super-resolution convolutional neural network. Journal of Measurement Science and Instrumentation， 2022， 13（3）： 276-283. DOI： 10. 3969／j．issn．1674-8042．2022．03．004
[full text view]
