An Enhanced Image Super-Resolution Via Generative Adversarial Network

Dongmei Ma ； Yu Li

《World Scientific Research Journal》 9卷3期 (2023/03) Pp. 98-107

https://doi.org/10.6911/WSRJ.202303_9(3).0012

摘要

Image super‐resolution algorithms have problems such as excessive smoothing of reconstructed images and poor visual quaity. Based on the existing image super‐ resolution algorithm SRGAN, an improved algorithm named PA‐SRGAN is proposed. The generator uses residual dense connection blocks instead of traditional residual units to ensure the input low‐resolution image features are fully transmitted throughout the generator, and pyramid attention is introduced to further improve the model performance. The discriminator adopts a relativistic average discriminator to improve the quality of the generative images and spectral normalization is used to enhance the training stability of the discriminant network. The whole model uses perceptual loss to prevent the image from being over‐smoothed and improve the perceptual quality of generated images. The experimental results show that the algorithm can obtain better perception coefficients in Set5, Set14, BSD100, Urban100 general test sets, the generative images are more in line with human visual perception.

關鍵字

Image super‐resolution ； Generatative adversarial network ； Residual dense connection ； Relativistic average discriminator ； Perceptual loss

參考文獻

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.

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.

Wang Xintao, Yu Ke, Dong Chao, et al. Recovering Realistic Texture in Image Super-Resolution by Deep Spatial Feature Transform[C]// Proc of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR).USA:IEEE, 2018:606-615.

Soh J W, Park G Y, Jo J, et al. Natural and realistic single image super-resolution with explicit natural maniford discrimination[C]//Proc of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR).USA:IEEE, 2019:8114-8123.

Zhang Yulun, Tian Yapeng, Kong Yu, et al. Residual dense network for image super-resolution[J].arXiv preprint arXiv:1802.08797,2018.