Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement
The deflection of face angle is the most important factor affecting the accuracy in face recognition, non-frontal faces make some face recognition systems lose their due functions, the existing frontal face conversion methods often have the phenomenon of distortion and lack of identity. Aiming at th...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10810390/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1823857119867174912 |
|---|---|
| author | Jihui Zhao Zhengyi Yuan Junhu Zhang Haitao Li Hui Li |
| author_facet | Jihui Zhao Zhengyi Yuan Junhu Zhang Haitao Li Hui Li |
| author_sort | Jihui Zhao |
| collection | DOAJ |
| description | The deflection of face angle is the most important factor affecting the accuracy in face recognition, non-frontal faces make some face recognition systems lose their due functions, the existing frontal face conversion methods often have the phenomenon of distortion and lack of identity. Aiming at the above problems, this paper proposes an face-frontal network which combines heat map of key points of face and improved attention mechanism. The network consists of a generator network and two discriminators networks, and the thermal map of the key points of the frontal face is used as a priori condition to guide the generation of the frontal face. In the generator part, the self-attention mechanism is introduced to obtain the dependence between feature points and other position features, which enhances the illumination perception ability of the network layer. At the same time, local attention is used in a discriminator to improve the local detail generation ability of the network in the face. Compared with other advanced frontal face generation methods, the proposed method has improved the accuracy of Rank-1 face recognition compared with other methods. The recognition rate of Rank-1 on Multi-PIE data set with small angle deflection is higher than other methods, the average recognition rate of Rank-1 on Multi-PIE setting2 is 97.41%, which is higher than the advanced method. Experimental results show that the proposed method can generate positive faces with corresponding identities from non-positive faces, which can be directly used in recognition tasks and has high recognition accuracy. |
| format | Article |
| id | doaj-art-32751dc20b5e496e8b51966c57992180 |
| institution | Kabale University |
| issn | 2169-3536 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IEEE |
| record_format | Article |
| series | IEEE Access |
| spelling | doaj-art-32751dc20b5e496e8b51966c579921802025-02-12T00:01:57ZengIEEEIEEE Access2169-35362025-01-0113249882499610.1109/ACCESS.2024.352061810810390Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism ImprovementJihui Zhao0https://orcid.org/0009-0005-8399-7284Zhengyi Yuan1https://orcid.org/0009-0009-3114-5484Junhu Zhang2Haitao Li3Hui Li4https://orcid.org/0000-0002-8533-2084School of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong, ChinaSchool of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong, ChinaSchool of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong, ChinaSchool of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong, ChinaSchool of Information Science and Technology, Qingdao University of Science and Technology, Qingdao, Shandong, ChinaThe deflection of face angle is the most important factor affecting the accuracy in face recognition, non-frontal faces make some face recognition systems lose their due functions, the existing frontal face conversion methods often have the phenomenon of distortion and lack of identity. Aiming at the above problems, this paper proposes an face-frontal network which combines heat map of key points of face and improved attention mechanism. The network consists of a generator network and two discriminators networks, and the thermal map of the key points of the frontal face is used as a priori condition to guide the generation of the frontal face. In the generator part, the self-attention mechanism is introduced to obtain the dependence between feature points and other position features, which enhances the illumination perception ability of the network layer. At the same time, local attention is used in a discriminator to improve the local detail generation ability of the network in the face. Compared with other advanced frontal face generation methods, the proposed method has improved the accuracy of Rank-1 face recognition compared with other methods. The recognition rate of Rank-1 on Multi-PIE data set with small angle deflection is higher than other methods, the average recognition rate of Rank-1 on Multi-PIE setting2 is 97.41%, which is higher than the advanced method. Experimental results show that the proposed method can generate positive faces with corresponding identities from non-positive faces, which can be directly used in recognition tasks and has high recognition accuracy.https://ieeexplore.ieee.org/document/10810390/Frontal facegeneration modelattention mechanismface identification |
| spellingShingle | Jihui Zhao Zhengyi Yuan Junhu Zhang Haitao Li Hui Li Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement IEEE Access Frontal face generation model attention mechanism face identification |
| title | Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement |
| title_full | Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement |
| title_fullStr | Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement |
| title_full_unstemmed | Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement |
| title_short | Frontal Face Generation Based on Attitude Point Guidance and Attention Mechanism Improvement |
| title_sort | frontal face generation based on attitude point guidance and attention mechanism improvement |
| topic | Frontal face generation model attention mechanism face identification |
| url | https://ieeexplore.ieee.org/document/10810390/ |
| work_keys_str_mv | AT jihuizhao frontalfacegenerationbasedonattitudepointguidanceandattentionmechanismimprovement AT zhengyiyuan frontalfacegenerationbasedonattitudepointguidanceandattentionmechanismimprovement AT junhuzhang frontalfacegenerationbasedonattitudepointguidanceandattentionmechanismimprovement AT haitaoli frontalfacegenerationbasedonattitudepointguidanceandattentionmechanismimprovement AT huili frontalfacegenerationbasedonattitudepointguidanceandattentionmechanismimprovement |