Abstract
Different face regions have different contributions to recognition. Especially in the wild environment, the difference of contributions will be further amplified due to a lot of interference. Based on this, this paper proposes an attention-aware face recognition method based on a deep convolutional neural network and reinforcement learning. The proposed method composes of an Attention-Net and a Feature-net. The Attention-Net is used to select patches in the input face image according to the facial landmarks and trained with reinforcement learning to maximize the recognition accuracy. The Feature-net is used for extracting discriminative embedding features. In addition, a regularization method has also been introduced. The mask of the input layer is also applied to the intermediate feature maps, which is an approximation to train a series of models for different face patches and provide a combined model. Our method achieves satisfactory recognition performance on its application to the public prevailing face verification database.
Original language | English |
---|---|
Pages (from-to) | 30-42 |
Number of pages | 13 |
Journal | IEEE Transactions on Biometrics, Behavior, and Identity Science |
Volume | 4 |
Issue number | 1 |
DOIs | |
State | Published - 1 Jan 2022 |
Keywords
- Attention-aware
- face recognition
- regularization
- reinforcement learning