Skip to main content
placeholder image

Face Verification Technology Based on FaceNet Similarity Recognition Network

Conference Paper


Abstract


  • The background complexity of face images is high in actual scenes, and there are a series of problems such as illumination and occlusion, which greatly reduces the performance of the face verification model. This paper proposes a face verification algorithm FaceNetSRM based on the FaceNet similarity recognition network to improve the performance of the face verification model and the accuracy of Chinese face verification. Firstly, the deep convolutional neural network framework in FaceNet is determined, and the similarity recognition module is used to replace the Euclidean distance module in FaceNet. Then, the CASIA-WebFace face dataset and the self-made face dataset C-facev1 are used to train the face verification algorithm of this article. Finally, the trained model is tested and evaluated on the face dataset LFW and CASIA-FaceV5 to show the effectiveness of the face verification method in this article, and the face verification effect of the algorithm is compared with the face verification effect of FaceNet. The experimental results show that the face verification accuracy rate of the FaceNetSRM algorithm in this paper is 1.5% higher than that of FaceNet, and the accuracy rate of Chinese face verification is improved by 2.8%. The algorithm has good robustness and generalization ability, which can be applied in face verification systems.

Publication Date


  • 2021

Citation


  • Gu, F., Lu, J., Xia, G., & Feng, Z. (2021). Face Verification Technology Based on FaceNet Similarity Recognition Network. In Proceedings of 2021 IEEE 10th Data Driven Control and Learning Systems Conference, DDCLS 2021 (pp. 1362-1367). doi:10.1109/DDCLS52934.2021.9455715

Scopus Eid


  • 2-s2.0-85114196061

Web Of Science Accession Number


Start Page


  • 1362

End Page


  • 1367

Abstract


  • The background complexity of face images is high in actual scenes, and there are a series of problems such as illumination and occlusion, which greatly reduces the performance of the face verification model. This paper proposes a face verification algorithm FaceNetSRM based on the FaceNet similarity recognition network to improve the performance of the face verification model and the accuracy of Chinese face verification. Firstly, the deep convolutional neural network framework in FaceNet is determined, and the similarity recognition module is used to replace the Euclidean distance module in FaceNet. Then, the CASIA-WebFace face dataset and the self-made face dataset C-facev1 are used to train the face verification algorithm of this article. Finally, the trained model is tested and evaluated on the face dataset LFW and CASIA-FaceV5 to show the effectiveness of the face verification method in this article, and the face verification effect of the algorithm is compared with the face verification effect of FaceNet. The experimental results show that the face verification accuracy rate of the FaceNetSRM algorithm in this paper is 1.5% higher than that of FaceNet, and the accuracy rate of Chinese face verification is improved by 2.8%. The algorithm has good robustness and generalization ability, which can be applied in face verification systems.

Publication Date


  • 2021

Citation


  • Gu, F., Lu, J., Xia, G., & Feng, Z. (2021). Face Verification Technology Based on FaceNet Similarity Recognition Network. In Proceedings of 2021 IEEE 10th Data Driven Control and Learning Systems Conference, DDCLS 2021 (pp. 1362-1367). doi:10.1109/DDCLS52934.2021.9455715

Scopus Eid


  • 2-s2.0-85114196061

Web Of Science Accession Number


Start Page


  • 1362

End Page


  • 1367