A discriminative self‐attention cycle GAN for face super‐resolution and recognition
Abstract Face image captured via surveillance videos in an open environment is usually of low quality, which seriously affects the visual quality and recognition accuracy. Most image super‐resolution methods adopt paired high‐quality and its interpolated low‐resolution version to train the super‐res...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2021-09-01
|
| Series: | IET Image Processing |
| Online Access: | https://doi.org/10.1049/ipr2.12250 |
| _version_ | 1828237102456766464 |
|---|---|
| author | Xiaoguang Li Ning Dong Jianglu Huang Li Zhuo Jiafeng Li |
| author_facet | Xiaoguang Li Ning Dong Jianglu Huang Li Zhuo Jiafeng Li |
| author_sort | Xiaoguang Li |
| collection | DOAJ |
| description | Abstract Face image captured via surveillance videos in an open environment is usually of low quality, which seriously affects the visual quality and recognition accuracy. Most image super‐resolution methods adopt paired high‐quality and its interpolated low‐resolution version to train the super‐resolution network. It is difficult to achieve contented visual quality and restoring discriminative features in real scenarios. A discriminative self‐attention cycle generative adversarial network is proposed for real‐world face image super‐resolution. Based on the cycle GAN framework, unpaired samples are adopted to train a degradation network and a reconstruction network simultaneously. A self‐attention mechanism is employed to capture the contextual information for details restoring. A Siamese face recognition network is introduced to provide a constraint on identify consistency. In addition, an asymmetric perceptual loss is introduced to handle the imbalance between the degradation model and the reconstruction model. Experimental results show that the observation model achieved more realistic low‐quality face images, and the super‐resolved face images have shown better subjective quality and higher face recognition performance. |
| first_indexed | 2024-04-12T20:44:03Z |
| format | Article |
| id | doaj.art-b901c72e90e14b97b61c4e171bee1ed6 |
| institution | Directory Open Access Journal |
| issn | 1751-9659 1751-9667 |
| language | English |
| last_indexed | 2024-04-12T20:44:03Z |
| publishDate | 2021-09-01 |
| publisher | Wiley |
| record_format | Article |
| series | IET Image Processing |
| spelling | doaj.art-b901c72e90e14b97b61c4e171bee1ed62022-12-22T03:17:20ZengWileyIET Image Processing1751-96591751-96672021-09-0115112614262810.1049/ipr2.12250A discriminative self‐attention cycle GAN for face super‐resolution and recognitionXiaoguang Li0Ning Dong1Jianglu Huang2Li Zhuo3Jiafeng Li4Faculty of Information Technology Beijing University of Technology Beijing ChinaFaculty of Information Technology Beijing University of Technology Beijing ChinaFaculty of Information Technology Beijing University of Technology Beijing ChinaFaculty of Information Technology Beijing University of Technology Beijing ChinaFaculty of Information Technology Beijing University of Technology Beijing ChinaAbstract Face image captured via surveillance videos in an open environment is usually of low quality, which seriously affects the visual quality and recognition accuracy. Most image super‐resolution methods adopt paired high‐quality and its interpolated low‐resolution version to train the super‐resolution network. It is difficult to achieve contented visual quality and restoring discriminative features in real scenarios. A discriminative self‐attention cycle generative adversarial network is proposed for real‐world face image super‐resolution. Based on the cycle GAN framework, unpaired samples are adopted to train a degradation network and a reconstruction network simultaneously. A self‐attention mechanism is employed to capture the contextual information for details restoring. A Siamese face recognition network is introduced to provide a constraint on identify consistency. In addition, an asymmetric perceptual loss is introduced to handle the imbalance between the degradation model and the reconstruction model. Experimental results show that the observation model achieved more realistic low‐quality face images, and the super‐resolved face images have shown better subjective quality and higher face recognition performance.https://doi.org/10.1049/ipr2.12250 |
| spellingShingle | Xiaoguang Li Ning Dong Jianglu Huang Li Zhuo Jiafeng Li A discriminative self‐attention cycle GAN for face super‐resolution and recognition IET Image Processing |
| title | A discriminative self‐attention cycle GAN for face super‐resolution and recognition |
| title_full | A discriminative self‐attention cycle GAN for face super‐resolution and recognition |
| title_fullStr | A discriminative self‐attention cycle GAN for face super‐resolution and recognition |
| title_full_unstemmed | A discriminative self‐attention cycle GAN for face super‐resolution and recognition |
| title_short | A discriminative self‐attention cycle GAN for face super‐resolution and recognition |
| title_sort | discriminative self attention cycle gan for face super resolution and recognition |
| url | https://doi.org/10.1049/ipr2.12250 |
| work_keys_str_mv | AT xiaoguangli adiscriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT ningdong adiscriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT jiangluhuang adiscriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT lizhuo adiscriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT jiafengli adiscriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT xiaoguangli discriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT ningdong discriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT jiangluhuang discriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT lizhuo discriminativeselfattentioncycleganforfacesuperresolutionandrecognition AT jiafengli discriminativeselfattentioncycleganforfacesuperresolutionandrecognition |