Image Super-Resolution Via Gradient Guidance and Gnenrative Adversarial Network

Dongmei Ma; Yu Li

doi:10.54097/jceim.v10i1.5074

Authors

Dongmei Ma
Yu Li

DOI:

https://doi.org/10.54097/jceim.v10i1.5074

Keywords:

Image super-resolution, Structural distortion, WGAN-GP, Gradient guidance, Adversarial training, Multi-scale residual block

Abstract

In order to solve the problem that the existing image super-resolution reconstruction model generates poor visual quality and prone to structural distortion, a deep gradient guidance based on generative adversarial network is proposed. The generator introduces the gradient branch to transfer the features of the gradient image and fuses the gradient information with the image branch to prevent the image edge distorted. Referring to MSRB, ResNext and Inception, an improved multi-scale residual block is proposed and applied to the basic module of the image branch and gradient branch, which makes model easier to obtain multi-scale information. The discriminator uses WGAN-GP to improve the stability of network training. Compared with the perceptual-driven algorithms such as SRGAN, ESRGAN and NatSR, the proposed algorithm can effectively prevent structural distortion of the image and improve the quality of the generated image. The computational complexity of proposed model is 23.7GFLOPs, which is lower than ESRGAN and SPSR about 1/4 and 1/10.

References

Tang Yanqiu, Pan Hong, Zhu Yaping, et al. A survey of image super-resolution reconstruction[J].Joural of electronic, 2020, 48(7):1407-1419.

Nan Fangzhe, Qian Yurong, Xing Yanni, et al.Survey of single image super-resolution based on deep learning[J]. Application Research of Computers, 2020, 37(2):321-326.

Dong Chao, Loy C C, He Kaiming, et al. Learning a deep convolutional network for image super-resolution[C] // Proc of the European Conference on Computer Vision (ECCV). Germany: Springer, 2014:184-199.

Kim J, Lee J K, Lee K M, et al. Accurate image super-resolution using very deep convolutional networks[C] // Proc of the Conference on Computer Vision and Pattern Recognition(CVPR). USA: IEEE，2016: 1646-1654．

Ledig C, Theis L, Huszár F, et al. Photo-realistic single image super-resolution using a generative adversarial network[C] // Proc of the Conference on Computer Vision and Pattern Recognition (CVPR). USA: IEEE, 2017:105-114.

Johnson J , Alahi A ,Fei-Fei L. Perceptual losses for real-time style transfer and super-resolution[J]. arXiv preprint arXiv:1603.08155, 2016.

Lim B, Son S, Kim H, et al. Enhanced deep residual networks for single image super-resolution[C]//Proc of the Conference on Computer Vision and Pattern Recognition Workshops (CVPRW). USA :IEEE, 2017:1132-1140.

Wang Xintao, Yu Ke, Wu Shixiang, et al. ESRGAN:Enhanced super-resolution generative adversarial network[C]//Proc of the European Conference on Computer Vision(ECCV). Germany: Springer, 2018:63-79.

Ma Cheng, Rao Yongming, Cheng Yean, et al. Structure-preserving super-resolution with gradient guidance[C]//Proc of the Conference on Computer Vision and Pattern Recognition (CVPR). USA :IEEE, 2020:7769-7778.

Hu J, Shen L, Albanie S, et al. Squeeze-and-excitation networks[J]. arXiv preprint arXiv:1709.01507, 2017.

Martin A, Soumith C, Léon B. Wasserstein GAN[J]. arXiv preprint arXiv:1701.07875, 2017.

Ishaan G, Faruk A, Martin A, et al. Improved training of Wasserstein GANs[J]. arXiv preprint arXiv:1704.00028v3, 2017.