Home // SIGNAL 2019, The Fourth International Conference on Advances in Signal, Image and Video Processing // View article

Residual Dense Generative Adversarial Network for Single Image Super-Resolution

Authors:
Jiahao Meng
Zekuan Yu
Tianping Shuai

Keywords: CNN; Single Image Super-Resolution; Generative Adversarial Networks.

Abstract:
Model-based very deep Convolutional Neural Networks (CNN) have achieved great success in Single Image Super-Resolution (SISR) work. However, most of the super-resolution models based on deep convolution networks can not fully utilize the hierarchical features of the original low-resolution images. In order to improve the quality of the high-frequency details of the reconstructed super-resolution image, we proposes a super-resolution method for Residual Dense Generative Adversarial Networks (RDGAN). We use the Generative Adversarial Networks (GAN) as our main model structure and the residual dense block as the basic building blocks of the generator, which makes the network pay more attention to the extraction of low-resolution image hierarchical features. Then, we fully exploit the hierarchical features from all the convolutional layers. Finally, we use perceptual loss as our loss function to get finer texture details and more realistic photo effects. Experiments show that our method can achieve significant improvement in the quality of high-frequency detail reconstruction at high magnification.

Pages: 17 to 22

Copyright: Copyright (c) IARIA, 2019

Publication date: June 2, 2019

Published in: conference

ISSN: 2519-8432

ISBN: 978-1-61208-716-0

Location: Athens, Greece

Dates: from June 2, 2019 to June 6, 2019