Abstract
Image super-resolution reconstruction (SR) aims to find the mapping relationship between a low-resolution (LR) image and the corresponding high-resolution (HR) image. At present, the SR methods based on deep learning still has the problem of blurred edges and loss of details in the reconstruction results. To address these issues, we propose an attention-based multi-scale SR network (AMSNet), In order to better learn the global and local features of images, a multi-scale network structure is designed to reconstruct richer image features. In the multi-scale structure, cascaded residual blocks (CRB) are used to extract image features. Additionally, a Squeeze-and-Excitation-based residual upsampling block (SERUB) is designed to enhance important features that are beneficial to image reconstruction. Extensive experiments show that our network has better performance in image reconstruction compared to some other state-of-the-art models and is able to reconstruct image texture details better.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have