A rapid face recognition BCI system using single-trial ERP

Face recognition has attracted great interest especially in computer vision, but poses and illumination of the face lay great difficulty in recognition accuracy. Studies on human face in the field of cognitive neuroscience found some face-specific event-related potential (ERP) components, such as N170, VPP. And some researchers used these face-specific ERP components to achieve face detection based Brain Computer Interface (BCI) system. In this paper, we introduce a BCI system using a rapid serial visual presentation (RSVP) paradigm to evaluate the effectiveness of face recognition with ERP components. We analyze the ERP components (N170, VPP, N2, P3) elicited by face recognition and evaluate their contributions to the rapid face recognition using single-trial ERP. We find that amplitudes of N170 and VPP show no significant difference between the target face and non-target face. The amplitude of N2 and P3 show significant difference. The averaged areas under ROC curve (AUC) of single-trial ERP classification reach 0.851 on rapid face recognition task for 8 subjects, and the best two reach 0.889 and 0.921. Our results show that the averaged AUC using N170 and VPP is 0.548, which indicates N170 and VPP are involved in the face recognition process. However, N2 and P3 contribute much more to the accuracy of the face recognition by single-trial ERP comparing to the trivial contributions of N170 and VPP.