| 基于人脸识别的人脸图像质量评估 |
投稿时间：2020-12-28 修订日期：2020-12-30  点此下载全文 |
| 引用本文： |
| 摘要点击次数: 881 |
| 全文下载次数: 0 |
|
|
| 中文摘要:无约束场景下，低质量的人脸图像不仅浪费计算资源而且降低系统识别率。针对此问题，提出一种基于人脸识别的人脸质量评估方法对人脸图像进行预评估。以人脸识别系统特征提取网络为基础网络在COX数据集上进行微调，并使用微调后网络对COX数据集进行质量分数标定。最后，结合基础网络及质量预测网络并以相应损失函数在标定数据上进行回归学习以获取质量评估模型。实验结果表明，该方法能够有效区分不同质量的人脸图像并提升人脸识别系统性能。 |
| 中文关键词:人脸质量评估 质量标定 人脸图像 人脸识别 |
| |
| Face Image Quality Assessment Based on Face Recognition |
|
|
| Abstract:In unconstrained scenarios, low quality face images not only waste computing resources but also reduce the recognition rate of the system. To solve this problem, a face quality assessment method based on face recognition is proposed to pre-evaluate face images. The feature extraction network in the face recognition system is used as the basic network and the COX dataset is fine-tuned. Then the COX dataset is annotated with the quality score by the fine-tuned network. Finally, the basic network and the quality prediction network are combined and the corresponding loss function is used to conduct regression learning on the labeled data to obtain the quality evaluation model. Experimental results show that this method can effectively distinguish different quality face images and improve the performance of face recognition system. |
| keywords:Face quality assessment Quality calibration Face image Face recognition |
| 查看全文 查看/发表评论 下载pdf阅读器 |
次