The Study Of EEG Signal In The Application Of Personal Identification

As biometrics being used more and more widely in our daily life, the demand for high security of personal identification system also raised. Although fingerprint, face recognition and other biometric system have been used widely, they have shortages of liveness detection, anti-spoofing, anti-forcing, etc.. In this paper, we studied the application of Electroencephalography (EEG) signal in personal identification. EEG is a validated effective biometric modality with distinct advantages. First, the active EEG must come from a living individual with a normal mental state. Second, EEG is hard to mimic. Finally, nowadays, advances in EEG recoding hardware dramatically simplified EEG acquisition procedure, and reduced the cost. In the past five years, EEG gradually emerged as a promising biometric modality to enhance the anti-spoofing capability of the existing biometric systems, and demonstrated some unique advantages in applications with high security requirements.In the second part of this thesis, pupil location in infrared video is realized. The goal is to locate the centre of pupil in infrared video, and it is of great value in practical applications. We present the location of pupil to BPPV (Benign Paroxysmal Positional Vertigo) diagnosis and treatment system. The system has been used in some hospitals now.The main tasks of this thesis include two parts:EEG-based personal identification and pupil location in infrared video. The details are as follows:(1) AR model parameters and power spectrum density are common features of EEG-based personal identification system. In this paper, we apply feature selection to the new EEG database, and extract Hilbert-Huang spectrum from EEG signal. What is more, the sparse representation-based classification to EEG-based personal identification system is also addressed.(2) We build an open-set personal identification system based on SVM. It solves the problem that k-NN classify cannot be applied to open-set situation. Experimental results show that EER is low and the accuracy is almost the same as k-NN classify on the same database.(3) Two dual-biometric-modality identification systems are implemented in this thesis. Fusion system makes advantages of fingerprint and EEG complementary to each other, the system of high accuracy and anti-spoofing, the performance of security was improved greatly.(4) Muscle signals always are treated as noise in EEG. However, we used muscle signals as a robust covert warning message novelty. Two algorithms based on time domain and Hilbert-Huang spectrum respectively are proposed, where both can detect covert warning messages perfectly. Experimental results verify the feasibility of EEG-based covert warning system.(5) A real time pupil location system is implemented with high accuracy, and balances the accuracy and efficiency perfectly. The median filter arithmetic, Hough transform circle detection arithmetic and Daugman’s integro-differential operator are improved to make the system efficient on the condition of high accuracy.