DFQA: Deep Face Image Quality Assessment
文献类型：会议论文
| 作者 | Yang, Fei1,2; Shao, Xiaohu1,2 ; Zhang, Lijun1,2; Deng, Pingling1,2; Zhou, Xiangdong1,2; Shi, Yu1,2
|
| 出版日期 | 2019 |
| 会议日期 | August 23, 2019 - August 25, 2019 |
| 会议地点 | Beijing, China |
| DOI | 10.1007/978-3-030-34110-7_55 |
| 页码 | 655-667 |
| 英文摘要 | A face image with high quality can be extracted dependable features for further evaluation, however, the one with low quality can’t. Different from the quality assessment algorithms for general images, the face image quality assessment need to consider more practical factors that directly affect the accuracy of face recognition, face verifcation, etc. In this paper, we present a two-stream convolutional neural network (CNN) named Deep Face Quality Assessment (DFQA) specifically for face image quality assessment. DFQA is able to predict the quality score of an input face image quickly and accurately. Specifcally, we design a network with two-stream for increasing the diversity and improving the accuracy of evaluation. Compared with other CNN network architectures and quality assessment methods for similar tasks, our model is smaller in size and faster in speed. In addition, we build a new dataset containing 3000 face images manually marked with objective quality scores. Experiments show that the performance of face recognition is improved by introducing our face image quality assessment algorithm. © 2019, Springer Nature Switzerland AG. |
| 会议录 | 10th International Conference on Image and Graphics, ICIG 2019
 |
| 语种 | 英语 |
| 电子版国际标准刊号 | 16113349 |
| ISSN号 | 03029743 |
| 源URL | [http://119.78.100.138/handle/2HOD01W0/9798]  |
| 专题 | 中国科学院重庆绿色智能技术研究院 |
| 作者单位 | 1.Chongqing Institute of Green and Intelligent Technology, CAS, Beijing; 400714, China; 2.University of Chinese Academy of Science, Beijing; 100049, China |
| 推荐引用方式 GB/T 7714 | Yang, Fei,Shao, Xiaohu,Zhang, Lijun,et al. DFQA: Deep Face Image Quality Assessment[C]. 见:. Beijing, China. August 23, 2019 - August 25, 2019. |
入库方式： OAI收割
来源：重庆绿色智能技术研究院
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。