A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution
Recently, deep convolutional neural networks (CNNs) have been widely explored in single image super-resolution(SISR) and obtained remarkable performance. However, most of the existing CNN-based SISR methods tend to produce over-smoothed outputs and miss some textural details. To address these issues...
Guardado en:
| Autores principales: | , , , , , , , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
IEEE
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/ffc04ca5aa0842d5bbeb8a176cd89c84 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| id |
oai:doaj.org-article:ffc04ca5aa0842d5bbeb8a176cd89c84 |
|---|---|
| record_format |
dspace |
| spelling |
oai:doaj.org-article:ffc04ca5aa0842d5bbeb8a176cd89c842021-11-19T00:06:23ZA Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution2169-353610.1109/ACCESS.2021.3058648https://doaj.org/article/ffc04ca5aa0842d5bbeb8a176cd89c842021-01-01T00:00:00Zhttps://ieeexplore.ieee.org/document/9352758/https://doaj.org/toc/2169-3536Recently, deep convolutional neural networks (CNNs) have been widely explored in single image super-resolution(SISR) and obtained remarkable performance. However, most of the existing CNN-based SISR methods tend to produce over-smoothed outputs and miss some textural details. To address these issues, we propose a wavelet-based asymmetric convolution network (WACN). Different from conventional CNN methods that directly infer HR images, our approach firstly learns to predict the LR’s corresponding series of HR’s wavelet coefficients before reconstructing HR images from them. This helps to capture more structural information in images to preserve texture information and avoid artifacts. To enhance the ability of feature extraction, we propose an asymmetric convolution block (ACB) structure to form a very deep network. In the training phase, ACB can provide different receptive fields to enrich feature information. In the inference phase, ACB’s asymmetric convolution kernel can be equivalently fused into the standard square-kernel layers, such that no extra computational burdens are introduced in the inference phase. Furthermore, we propose a variance-based channel attention (VCA) mechanism to adaptively rescale channel-wise features by considering interdependencies among channels. Extensive experimental results demonstrate the superiority of the proposed WACN in comparison with the state-of-the-art methods.Wanxu ZhangKai JiangLin WangNa MengYan ZhouYanyan LiHailong HuXiaoxuan ChenBo JiangIEEEarticleAsymmetric convolution networksingle super-resolutionvariance-based channel attentionwaveletElectrical engineering. Electronics. Nuclear engineeringTK1-9971ENIEEE Access, Vol 9, Pp 28976-28986 (2021) |
| institution |
DOAJ |
| collection |
DOAJ |
| language |
EN |
| topic |
Asymmetric convolution network single super-resolution variance-based channel attention wavelet Electrical engineering. Electronics. Nuclear engineering TK1-9971 |
| spellingShingle |
Asymmetric convolution network single super-resolution variance-based channel attention wavelet Electrical engineering. Electronics. Nuclear engineering TK1-9971 Wanxu Zhang Kai Jiang Lin Wang Na Meng Yan Zhou Yanyan Li Hailong Hu Xiaoxuan Chen Bo Jiang A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution |
| description |
Recently, deep convolutional neural networks (CNNs) have been widely explored in single image super-resolution(SISR) and obtained remarkable performance. However, most of the existing CNN-based SISR methods tend to produce over-smoothed outputs and miss some textural details. To address these issues, we propose a wavelet-based asymmetric convolution network (WACN). Different from conventional CNN methods that directly infer HR images, our approach firstly learns to predict the LR’s corresponding series of HR’s wavelet coefficients before reconstructing HR images from them. This helps to capture more structural information in images to preserve texture information and avoid artifacts. To enhance the ability of feature extraction, we propose an asymmetric convolution block (ACB) structure to form a very deep network. In the training phase, ACB can provide different receptive fields to enrich feature information. In the inference phase, ACB’s asymmetric convolution kernel can be equivalently fused into the standard square-kernel layers, such that no extra computational burdens are introduced in the inference phase. Furthermore, we propose a variance-based channel attention (VCA) mechanism to adaptively rescale channel-wise features by considering interdependencies among channels. Extensive experimental results demonstrate the superiority of the proposed WACN in comparison with the state-of-the-art methods. |
| format |
article |
| author |
Wanxu Zhang Kai Jiang Lin Wang Na Meng Yan Zhou Yanyan Li Hailong Hu Xiaoxuan Chen Bo Jiang |
| author_facet |
Wanxu Zhang Kai Jiang Lin Wang Na Meng Yan Zhou Yanyan Li Hailong Hu Xiaoxuan Chen Bo Jiang |
| author_sort |
Wanxu Zhang |
| title |
A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution |
| title_short |
A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution |
| title_full |
A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution |
| title_fullStr |
A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution |
| title_full_unstemmed |
A Wavelet-Based Asymmetric Convolution Network for Single Image Super-Resolution |
| title_sort |
wavelet-based asymmetric convolution network for single image super-resolution |
| publisher |
IEEE |
| publishDate |
2021 |
| url |
https://doaj.org/article/ffc04ca5aa0842d5bbeb8a176cd89c84 |
| work_keys_str_mv |
AT wanxuzhang awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT kaijiang awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT linwang awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT nameng awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT yanzhou awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT yanyanli awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT hailonghu awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT xiaoxuanchen awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT bojiang awaveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT wanxuzhang waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT kaijiang waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT linwang waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT nameng waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT yanzhou waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT yanyanli waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT hailonghu waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT xiaoxuanchen waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution AT bojiang waveletbasedasymmetricconvolutionnetworkforsingleimagesuperresolution |
| _version_ |
1718420647406731264 |