Fast and Reliable P300-Based BCI with Facial Images Akinari Onishi 1,3 , Yu Zhang 2,3 , Qibin Zhao 3 , Andrzej Cichocki 3 1 Department of Brain Science and Engineering, Kyushu Institute of Technology, Fukuoka, Japan 2 School of Information Science and Engineering, East China University of Science and Technology, Shanghai, China 3 Lab. for Advanced Brain Signal Processing, Brain Science Institute, Riken, Saitama, Japan onishi@brain.riken.jp Abstract A P300-based brain-computer interface (BCI) often called ”P300 speller” is a promising approach to help disabled people communicate with external world. However, the spellers using letters or symbols as stimuli usually require more than 5 repetitions to achieve high classification accuracy due to weak P300 evoked potential. We propose a suitable platform which has 8 command intuitive interface and uses human’s facial images as flashes. The experimental results from on-line tests demonstrated that our P300-based BCI with facial images achieved over 90% accuracies by using only two repetitions. We also analyzed off-line EEG data using two or three channels and 2 paradigms achieved over 80% accuracies. In addition, we found that the latencies of event-related potentials evoked by a clockwise flash order were shorter than that evoked by a random flash order. 1 Introduction An electroencephalographic (EEG) brain-computer interface (BCI) can provide a non-invasive, low cost, non-muscular means of direct communication between a human brain and a computer or a robot [1, 2]. One of the most reliable and promising multi-command BCI system is based on P300 evoked potential paradigm. Most of P300-based BCI platforms exploit so called P300 speller with small abstract elements like letters and symbols arranged in the form of a matrix where each row and column is intensified in a random sequence [1]. Recently, big progresses have been made in optimization of the P300 speller, for example, on color and intensity of stimulation [3], size of the matrix [4], EEG electrode locations [5] and classification algorithms [6]. Most of the spellers using letters or symbols as stimuli require more than 5 repetitions to reliably extract P300 evoked potential since the P300 evoked potential is relatively weak and occurs amid other ongoing EEG activities. Our objective in this paper is to demonstrate how to enhance or increase P300 evoked potentials by suitably designed stimuli. We present a novel P300-based BCI whose commands are intensified by natural images of human faces – the affective face driven paradigm (AFDP). In this paper we use P300-based BCI that has 8 independent commands to navigate, for example, a robot arm or a wheel chair moving in 8 different directions. Our experiments demonstrated that P300-based BCI using AFDP showed higher accuracies than that a gray/white flash of abstract symbols. 2 Methods We tested 6 paradigms illustrated in Figure 1. The central colored arrow image shows a target or an output as a feedback. The other arrow images indicate control commands and the part is intensified sequentially during each trial. We prepared two different flash orders: random (R) 1