Name
Abstract | ||
|---|---|---|
Because convolutional neural networks can extract discriminative features, they are widely used in face recognition and significantly improve the performance in face recognition. In order to improve the accuracy of the face recognition, in addition to improving the structures of convolutional neural networks, many new loss functions have been proposed to enhance the distinguishing ability to extract features, such as SphereFace and ArcFace. Inspired by SphereFace and ArcFace, we propose a new loss function called MaaFace, in which the angular multiplier and the angular addition are introduced into the loss function simultaneously. We give a detailed derivation of MaaFace and conduct extensive experiments on different networks and different data sets. Experiments show that our proposed loss function can achieve an out performance than the latest face recognition loss functions in face recognition accuracy. Finally, we explain why MaaFace can achieve better performance through statistical analysis of the experimental data. |
Year | DOI | Venue |
|---|---|---|
2019 | 10.1007/978-3-030-34113-8_53 | IMAGE AND GRAPHICS, ICIG 2019, PT III |
Keywords | DocType | Volume |
Face recognition, Convolutional neural networks, Loss function | Conference | 11903 |
ISSN | Citations | PageRank |
0302-9743 | 0 | 0.34 |
References | Authors | |
0 | 5 | |
Name | Order | Citations | PageRank |
|---|---|---|---|
| Weilun Liu | 1 | 2 | 1.38 |
| Jichao Jiao | 2 | 18 | 6.53 |
| Yaokai Mo | 3 | 2 | 1.72 |
| Jian Jiao | 4 | 2 | 2.06 |
| Zhongliang Deng | 5 | 0 | 1.35 |