中国邮电高校学报(英文) ›› 2018, Vol. 25 ›› Issue (1): 54-61.doi: 10.19682/j.cnki.1005-8885.2018.0006

• Artificial Intelligence • 上一篇    下一篇

Learning an identity distinguishable space for large scale face recognition

岳婷1,王洪波1,程时端2   

  1. 1. 北京邮电大学
    2. 北京邮电大学网络与交换技术国家重点实验室
  • 收稿日期:2017-07-17 修回日期:2018-01-15 出版日期:2018-02-28 发布日期:2018-02-28
  • 通讯作者: 王洪波 E-mail:hbwang@bupt.edu.cn
  • 基金资助:
    国家863计划;国家自然科学基金;中央高校基本科研业务费专项基金

Learning an identity distinguishable space for large scale face recognition

  • Received:2017-07-17 Revised:2018-01-15 Online:2018-02-28 Published:2018-02-28
  • Contact: Hong-Bo WANG E-mail:hbwang@bupt.edu.cn
  • Supported by:
    ;Natural Science Foundation of China

摘要: Implementing face recognition efficiently to real world large scale dataset presents great challenges to existing approaches.The method in this paper was proposed to learn an identity distinguishable space for large scale face recognition in MSR- Bing image recognition challenge (IRC). Firstly, a deep convolutional neural network (CNN) was used to optimize a 128 B embedding for large scale face retrieval. The embedding was trained via using triplets of aligned face patches from FaceScrub and CASIA-WebFace datasets. Secondly, the evaluation of MSR-Bing IRC was conducted according to a cross-domain retrieval scheme. The real-time retrieval in this paper was benefited from the K-means clustering performed on the feature space of training data. Furthermore, a large scale similarity learning (LSSL) was applied on the relevant face images for learning a better identity space. A novel method for selecting similar pairs was proposed for LSSL. Compared with many existing networks of face recognition, the proposed model was lightweight and the retrieval method was promising as well.

关键词:

Abstract: Implementing face recognition efficiently to real world large scale dataset presents great challenges to existing approaches.The method in this paper was proposed to learn an identity distinguishable space for large scale face recognition in MSR- Bing image recognition challenge (IRC). Firstly, a deep convolutional neural network (CNN) was used to optimize a 128 B embedding for large scale face retrieval. The embedding was trained via using triplets of aligned face patches from FaceScrub and CASIA-WebFace datasets. Secondly, the evaluation of MSR-Bing IRC was conducted according to a cross-domain retrieval scheme. The real-time retrieval in this paper was benefited from the K-means clustering performed on the feature space of training data. Furthermore, a large scale similarity learning (LSSL) was applied on the relevant face images for learning a better identity space. A novel method for selecting similar pairs was proposed for LSSL. Compared with many existing networks of face recognition, the proposed model was lightweight and the retrieval method was promising as well.

Key words: face recognition, convolutional neural network (CNN), triplet loss, inception, similarity learning