A general lightweight image super-resolution with sharpening enhancement and double attention network

In recent years, single image super-resolution (SISR) based on deep learning has achieved excellent results. However, the consequent elevated computational and storage expenses limit its practicability in real life. Researchers seek a lightweight SISR network that minimizes computational load while maintaining high performance. To address this challenge, we introduce a general lightweight image super-resolution with sharpening enhancement and double attention network (ESDAN) to optimize the trade-off between model complexity and performance. The network achieves a balance between model complexity and performance through the Sharpening Enhancement Module (SEM) and the Dual Attention Upsampling module (DAU). Specifically, SEM effectively integrates the Attention-Driven Feature Sharpening module (ADFS) to enhance feature contrast and the Multi-Way Feature Enhancement module (MWFE) to reinforce key information, optimizing both the representation ability of composite features and the nonlinear mapping ability of the model. Moreover, DAU dynamically fuses shallow and deep features to enhance the model's reconstruction capability. Extensive experimental results demonstrate that the proposed network surpasses contemporary state-of-the-art lightweight SISR methods. Additionally, we explore the potential of ESDAN in other SISR-related tasks, such as super-resolution of Alzheimer's disease brain MRI, stereo endoscopic images, and surveillance images. The experimental results demonstrate the high versatility of the proposed network. The source code is available at https://github.com/Czs138/ESDAN .