Year: 2025
Type: article
Abstract: Image super-resolution (SR) is a fundamental challenge in the field of computer vision. Recently, Convolutional Neural Network (CNN)-based methods for image SR have achieved significant progress across various SR tasks. However, most current research focuses on designing deeper and wider architectures, often sacrificing computational burden and speed in order to improve image SR quality. To achieve more efficient SR methods, this paper proposes a Wavelet Attention Network (WANet) for image SR. Firstly, a wavelet-based attention module is proposed. Compared to existing self-attention modules, the wavelet attention module decomposes image features into different frequency components using wavelet transforms. It then applies a self-attention mechanism to capture multi-scale features, enabling a more efficient and larger receptive field to help the network capture long-range feature dependencies. Secondly, local implicit features are introduced to enhance the encoder’s ability to aggregate local neighborhood features. Finally, coarse and fine-grained interwoven pixel features are collaboratively associated to improve the performance of the implicit feature decoder. Experimental comparisons with state-of-the-art SR methods demonstrate the effectiveness and superiority of WANet in the field of image SR.
Source: Complex & Intelligent Systems
Authors: Chen Boying, Shi Jie
Institution: Suzhou University of Science and Technology
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Estimated APC paid: $2,490
Award: G1066389045
OpenAlex ID: W4409370483