A complete and fully automated face verification system on mobile devices

• Published in: Pattern recognition Vol. 46; no. 1; pp. 45 - 56
• Main Authors: Ren, Jianfeng, Jiang, Xudong, Yuan, Junsong
• Format: Journal Article
• Language: English
• Published: Kidlington Elsevier Ltd 01.01.2013; Elsevier
• Subjects: Applied sciences; Attentional cascade; Automated; Cascades; Class-specific threshold; Detectors; Exact sciences and technology; Eye detection; Eyes; Face detection; Face verification system; Image processing; Information, signal and communications theory; Mobile communication systems; Pattern recognition; Security; Signal processing; Subspace approach; Telecommunications and information theory; Attentional cascade; Face verification system; Class-specific threshold; Face detection; Eye detection; Subspace approach; Biometrics; Performance evaluation; Face recognition; Image processing; Error rate; Subspace method; Pattern recognition; Implementation; Learning; Subsystem; Database; Open market; Safety; Automatic recognition; Learning algorithm
• ISSN: 0031-3203; 1873-5142
• DOI: 10.1016/j.patcog.2012.06.013

Mobile devices have been widely used not only as a communication tool, but also a digital assistance to our daily life, which imposes high security concern on mobile devices. In this paper we present a natural and non-intrusive way to secure mobile devices, i.e. a complete and fully automated face verification system. It consists of three sub-systems: face detection, alignment and verification. The proposed subspace face/eye detector locates the eyes at a much higher precision than Adaboost face/eye detector. By utilizing attentional cascade strategy, the proposed face/eye detector achieves a comparable speed to Adaboost face/eye detector in this “close-range” application. The proposed approach that determines the class-specific threshold without sacrificing the training data for the validation data further boosts the performance. The proposed system is systematically evaluated on O2FN, AR and CAS-PEAL databases, and compared with many different approaches. Compared to the best competitive system, which is built upon Adaboost face/eye detector and ERE approach for face recognition, the proposed system reduces the overall equal error rate from 8.49% to 3.88% on the O2FN database, from 7.64% to 1.90% on the AR database and from 9.30% to 5.60% on the CAS-PEAL database. The proposed system is implemented on O2 XDA Flame and on average it takes 1.03s for the whole process, including face detection, eye detection and face verification. ► Complete and fully automated face verification system on mobile devices is proposed. ► The proposed system greatly reduces equal error rate of face recognition system. ► Cascade-APCDA detector is more accurate than Adaboost detector. ► Class-specific threshold based on training data is better than global threshold.