Joint restoration convolutional neural network for low-quality image super resolution
- PDF / 3,058,138 Bytes
- 20 Pages / 595.276 x 790.866 pts Page_size
- 14 Downloads / 173 Views
ORIGINAL ARTICLE
Joint restoration convolutional neural network for low-quality image super resolution Gadipudi Amaranageswarao1
· S. Deivalakshmi1 · Seok-Bum Ko2
Accepted: 13 October 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract In this paper, a joint restoration convolutional neural network (JRCNN) is proposed to produce a visually pleasing super resolution (SR) image from a single low-quality (LQ) image. The LQ image is a low resolution (LR) image with ringing, blocking and blurring artifacts arising due to compression. JRCNN consists of three deep dense residual blocks (DRB). Each DRB comprises of parallel convolutional layers with cross residual connections. The representational power of JRCNN is improved by depth-wise concatenation of feature representations from each of the DRBs. Moreover, these connections mitigate the problem of vanishing of gradients. Different from the previous networks, JRCNN exploits the contextual information directly in the LR image space without using any interpolation. This strategy improves the training efficiency of the network. The exhaustive experimentation on different datasets show that the proposed JRCNN produces state-of-the-art performance. Furthermore, ablation experiments are performed to assess the effectiveness of JRCNN. In addition, individual experiments are conducted for SR and compression artifact removal on benchmark datasets. Keywords Blocking artifacts · Cross residual connections · Dense residual blocks · Ringing · Skip connections
1 Introduction In this era of information explosion, transfer of multimedia content among different networks and devices have become more popular. Compression and down sampling are the most common image degradations. Images are available in degraded form on the web due to the availability of low storage space. Furthermore, images are compressed while being transmitted through low bandwidth channels. Down sampling exploits the spatial redundancy (duplicate pixels) in an image while compression further exploits the correlation in frequency (DCT coefficients) and temporal domains for frames in a video. Even though the degraded image requires less storage space, it contains unpleasant visual artifacts, i.e., ringing, blocking and blurring. These complex visual artifacts severely affect the user experience (e.g., Fig. 1). Accurate reconstruction of the super resolution (SR) image
B
Gadipudi Amaranageswarao [email protected]
1
Department of Electronics and Communication Engineering, National Institute of Technology, Thiruchirappalli, India
2
Department of Electrical and Computer Engineering, University of Saskatchewan, Saskatoon, Canada
from a single LQ image is a challenging task. There is a huge requirement for improvement in visual quality of reconstructed SR image. The lossy Joint Photographic Experts Group (JPEG) compression [32] standard introduces blocking artifacts (e.g., Fig. 2) due to discontinuities between the adjacent 8×8 pixel blocks, while the ringing and blurring artifacts occur du
Data Loading...
