SRCAN: Symmetric Residual Channel Attention Network for Efficient Single Image Super-Resolution
Year: 2026
Type: conference-paper
Abstract: With the advancement of electronic information systems, image information is indispensable in daily life and national science and technology, yet practical images often suffer from blurring. Due to the limitations and costs of hardware upgrades, software-based super-resolution (SR) has become a mainstream alternative. This paper proposes a Symmetric Residual Channel Attention Network (SRCAN): it deepens the network for detail extraction, integrates symmetric residual learning to mitigate information loss in deep networks, and adopts a channel attention mechanism for adaptive feature learning. Deconvolution layers replace interpolation for reconstruction to better recover image details. After testing on the public dataset, the data indicated that SRCAN performed better than the previous version: the Peak Signal-to-Noise Ratio (PSNR) was improved, the Structural Similarity (SSIM) remained at a high level, while the Learned Perceptual Image Patch Similarity (LPIPS) decreased.
Authors: Shuyi Cao, Yuying Wang, Rongcheng Zhang
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204176008