An Improvement of the VDSR Network for Single Image Super-Resolution by Truncation and Adjustment of the Learning Rate Parameters

Vadim Romanuke

doi:10.2478/acss-2019-0008

Abstract

A problem of single image super-resolution is considered, where the goal is to recover one high-resolution image from one low-resolution image. Whereas this problem has been successfully solved so far by the known VDSR network, such an approach still cannot give an overall beneficial effect compared to bicubic interpolation. This is so due to the fact that the image reconstruction quality has been estimated separately by three subjective factors. Moreover, the original VDSR network consisting of 20 convolutional layers is apparently not optimal by its depth. This is why here those factors are aggregated, and the network performance is deemed by a single estimator. Then the depth is tried to be decreased (truncation) along with adjusting the learning rate drop factor. Finally, a plausible improvement of the VDSR network is confirmed. The best truncated network, performing by almost 3.2 % better than bicubic interpolation, occupies less memory space and is about 1.44 times faster than the original VDSR network for images of a medium size.

References

[1] J. Salvador, Example-Based Super Resolution. Academic Press, 2017, 162 p. https://doi.org/10.1016/C2015-0-06719-310.1016/C2015-0-06719-3
Search in Google Scholar Back to article
[2] V. Bannore, Iterative-Interpolation Super-Resolution Image Reconstruction. A Computationally Efficient Technique. Springer-Verlag Berlin Heidelberg, 2009, 113 p. https://doi.org/10.1007/978-3-642-00385-110.1007/978-3-642-00385-1
Search in Google Scholar Back to article
[3] A. Small and S. Stahlheber, “The Role of Image Analysis Algorithms in Super-resolution Localization Microscopy,” in: Fluorescence Microscopy, Cornea A., Conn P. M. (eds.). Academic Press, 2014, pp. 227–242. https://doi.org/10.1016/B978-0-12-409513-7.00016-610.1016/B978-0-12-409513-7.00016-6
Search in Google Scholar Back to article
[4] M. Amiri, A. Ahmadyfard, and V. Abolghasemi, “A fast video super resolution for facial image,” Signal Processing: Image Communication, vol. 70, 2019, pp. 259–270. https://doi.org/10.1016/j.image.2018.10.00810.1016/j.image.2018.10.008
Search in Google Scholar Back to article
[5] R. Guo, X. Shi, Y. Zhu, and T. Yu, “Super-resolution reconstruction of astronomical images using time-scale adaptive normalized convolution,” Chinese Journal of Aeronautics, vol. 31, no. 8, pp. 1752–1763, 2018. https://doi.org/10.1016/j.cja.2018.06.00210.1016/j.cja.2018.06.002
Search in Google Scholar Back to article
[6] I. Haq and A. A. Mudassar, “Geometric super-resolution using negative rect mask,” Optik, vol. 168, pp. 323–341, 2018. https://doi.org/10.1016/j.ijleo.2018.04.03310.1016/j.ijleo.2018.04.033
Search in Google Scholar Back to article
[7] K. Hayat, “Multimedia super-resolution via deep learning: A survey,” Digital Signal Processing, vol. 81, pp. 198–217, 2018. https://doi.org/10.1016/j.dsp.2018.07.00510.1016/j.dsp.2018.07.005
Search in Google Scholar Back to article
[8] J. Lu, W. Hu, and Y. Sun, “A deep learning method for image super-resolution based on geometric similarity,” Signal Processing: Image Communication, vol. 70, pp. 210–219, 2019. https://doi.org/10.1016/j.image.2018.10.00310.1016/j.image.2018.10.003
Search in Google Scholar Back to article
[9] J. Kim, J. K. Lee, and K. M. Lee, “Accurate image super-resolution using very deep convolutional networks,” 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 1646–1654, 2016. https://doi.org/10.1109/CVPR.2016.18210.1109/CVPR.2016.182
Search in Google Scholar Back to article
[10] W. Xie, Y. Li, and X. Jia, “Deep convolutional networks with residual learning for accurate spectral-spatial denoising,” Neurocomputing, vol. 312, pp. 372–381, 2018. https://doi.org/10.1016/j.neucom.2018.05.11510.1016/j.neucom.2018.05.115
Search in Google Scholar Back to article
[11] R. Kress, “Interpolation,” in: Numerical Analysis, Kress R. (ed.). Springer, 1998, pp. 151–188. https://doi.org/10.1007/978-1-4612-0599-9_810.1007/978-1-4612-0599-9_8
Search in Google Scholar Back to article
[12] F. Aràndiga, “A nonlinear algorithm for monotone piecewise bicubic interpolation,” Applied Mathematics and Computation, vol. 272, no. 1, pp. 100–113, 2016. https://doi.org/10.1016/j.amc.2015.08.02710.1016/j.amc.2015.08.027
Search in Google Scholar Back to article
[13] A. M. Bayen and T. Siauw, “Interpolation,” in: An Introduction to MATLAB^® Programming and Numerical Methods for Engineers, Bayen A. M., Siauw T. (eds.). Academic Press, 2015, pp. 211–223. https://doi.org/10.1016/B978-0-12-420228-3.00014-210.1016/B978-0-12-420228-3.00014-2
Search in Google Scholar Back to article
[14] D. Brunet, J. Vass, E. R. Vrscay, and Z. Wang, “Geodesics of the structural similarity index,” Applied Mathematics Letters, vol. 25, no. 11, pp. 1921–1925, 2012. https://doi.org/10.1016/j.aml.2012.03.00110.1016/j.aml.2012.03.001
Search in Google Scholar Back to article
[15] A. Mittal, R. Soundararajan, and A. C. Bovik, “Making a “completely blind” image quality analyzer,” IEEE Signal Processing Letters, vol. 20, no. 3, pp. 209–212, 2013. https://doi.org/10.1109/LSP.2012.222772610.1109/LSP.2012.2227726
Search in Google Scholar Back to article
[16] L. Zhou, Z. Wang, S. Wang, and Y. Luo, “Coarse-to-Fine Image Super-Resolution Using Convolutional Neural Networks,” in: MultiMedia Modeling. MMM 2018. Lecture Notes in Computer Science, Schoeffmann K. et al. (eds.). Springer, 2018, pp. 73–81. https://doi.org/10.1007/978-3-319-73600-6_710.1007/978-3-319-73600-6_7
Search in Google Scholar Back to article
[17] V. V. Romanuke, “Appropriateness of numbers of receptive fields in convolutional neural networks based on classifying CIFAR-10 and EEACL26 datasets,” Electrical, Control and Communication Engineering, vol. 14, no. 2, pp. 157–163, 2018. https://doi.org/10.2478/ecce-2018-001910.2478/ecce-2018-0019
Search in Google Scholar Back to article

An Improvement of the VDSR Network for Single Image Super-Resolution by Truncation and Adjustment of the Learning Rate Parameters

Abstract

Paradigm

My account