Using Brain Waves as New Biometric Feature for Authenticating a Computer User in Real-Time
Download as PPTX, PDF3 likes631 views
This document presents a brain-computer interface (BCI) for person authentication using electroencephalography (EEG) signals. The BCI is used to lock and unlock a computer screen based on a user's brain activity. EEG data is collected from 14 sensors while users perform mental tasks. Power spectral density is calculated from the EEG signals and used as features for a two-stage classification system. The system achieves over 78% accuracy in authenticating users based on their brain activity patterns.
Using Brain Waves as New Biometric Feature for Authenticating a Computer User in Real-Time
- 1. International Journal of Biometrics and Bioinformatics (IJBB), Volume (7) : Issue (1) : 2013 Presented by buthainah hamdy
- 2. Abstract 1-INTRODUCTION 2-RELATED WORK 3-METHODOLOGY 3-1 Data Acquisition 3-2 Preprocessing and Feature Extraction 3-3 Classification 3-4 Implementation 4-CONCLUSION
- 3. We propose an EEG based BCI as new modality for Person Authentication an develop a screen lock application that will lock and unlock the computer screen. The brain waves of the person, recorded in real time are used as password to unlock the screen. Using 14 sensors of Emotiv headset. Compute the power spectral density. The channel spectral power in the frequency band of alpha, beta and gamma is used in the classification task.
- 4. A two stage checking is done to authenticate the user. A proximity value of 0.78 and above is considered a good match. The percentage of accuracy in classification is found to be good. No external stimulus is used.
- 5. Abstract 1-INTRODUCTION 2-RELATED WORK 3-METHODOLOGY 3-1 Data Acquisition 3-2 Preprocessing and Feature Extraction 3-3 Classification 3-4 Implementation 4-CONCLUSION
- 6. In this computer driven era, with the increase in security threats, securing and managing the resources has become a more complex challenge. Therefore, it is crucial to design a high security system for authentication. The world getting ready to transit from Graphic User Interface (GUI) to Natural User Interface (NUI) technology. We have made an attempt to build an authentication system based on thoughts.
- 7. Are based on personal identification number (PIN) and password that can be attacked by “shoulder surfing”. The biometric approaches based on the biological characteristics of humans cannot be hacked, stolen or transferred from one person to another as they are unique for each person. But it change with age and time.
- 8. Multimodal fusion for identity verification has shown great improvement compared to unimodal algorithms where they propose to integrate confidence measures during the fusion process.
- 9. Mostly use fingerprints, speech, facial features, iris and signatures as a base for an authentication or an identification system. These traits however, are known to be vulnerable to falsification as it is possible to forge or steal. Therefore, new types of physiological features that are unique and cannot be replicated are proposed for an identification system. Electroencephalogram (EEG) signal as a biometric. The EEG based biometrics is widely being considered in security sensitive areas like banks, labs and identification of criminal in forensic. It can be used as a component of National e-identity card in government sector, as they have proven to be unique between people.
- 10. Aims to convey people's intentions to the outside world directly from their thoughts, and is a direct communication pathway between a brain and an external device. A common method for designing BCI is to use EEG signals extracted during mental tasks that brain records using EEG sensors (electrodes). Person authentication aims to accept or reject a person claiming an identity, comparing a biometric data to one template, while the goal of person identification is to match the biometric data against all the records in a database. In our work, we have made an attempt to authenticate a system, rather than identification. Brain waves measured by EEG represent a summary of brain electrical activity at a recording point on the scalp . The fusion of delta, theta, alpha/mu, beta and gamma waves in frequency band.
- 11. Illustration of Location of Electrodes on the Emotiv Headset
- 12. We perform data acquisition, feature extraction, matching the feature vector with the stored template all in real time. we have used Power Spectral Density(PSD) as a reliable feature. PSD is the measure of the how much power strength at each frequency. It shows which frequencies variations are strong and which frequencies variations are weak. Principal Component Analysis(PCA) is applied to reduce the feature size. The obtained feature vector is then compared against a previously stored feature vector for the same person using template matching.
- 13. The match is considered good if the result of the comparison is greater than the threshold value 0.78 after repeated trials keeping in mind the need to satisfy low False Acceptance Error (FAE) and False Rejection Error (FRE). False Rejections(FR) will be approximately equal to the proportion of False Acceptances (FA)called as Equal Error Rate(EER).
- 14. We have developed a GUI, to let a user lock his computer screen when required and unlock the same by recording his brain activity (EEG signals) as a password for the system. An identity authentication system has to deal with two kinds of events: 1-either the person claiming a given identity is the one who he claims to be (in which case, he is called a client) 2-he is not(in which case, he is called an impostor). The main aim is to keep the False Acceptance Error (FAE) and the False Rejection Error (FRE) close to zero.
- 15. Abstract 1-INTRODUCTION 2-RELATED WORK 3-METHODOLOGY 3-1 Data Acquisition 3-2 Preprocessing and Feature Extraction 3-3 Classification 3-4 Implementation 4-CONCLUSION
- 16. EEG based person authentication was first proposed by Marcel. Used Power Spectral Density as the feature ,a statistical framework based on Gaussian Mixture Models (GMM) and Maximum A Posteriori Model (MAP) Adaptation on speaker and face authentication. Neural network classification was performed on real EEG data of healthy individuals to experimentally investigate the connection between a person's EEG and genetically specific information. correct classification scores in the range of 80% to 100% for person identification.
- 17. Two-stage biometric authentication method was proposed. The feature extraction methodology includes both linear and nonlinear measures to give improved accuracy. The combination of two-stage authentication with EEG features has good potential as a biometric as it is highly resistant to fraud. Principal Component Analysis(PCA) is applied to reduce the feature size.
- 18. Abstract 1-INTRODUCTION 2-RELATED WORK 3-METHODOLOGY 3-1 Data Acquisition 3-2 Preprocessing and Feature Extraction 3-3 Classification 3-4 Implementation 4-CONCLUSION
- 19. Framework of the Model.
- 20. EEG signals are recorded. The sampling rate is 128Hz. The total time of each recording is 10 seconds. The subject is instructed to avoid blinking or moving his body during the data collection to prevent the noise caused due to artifacts. Artifacts due to eye blinks produces a high amplitude signal called Electrooculogram (EOG) that can be many times greater than the EEG signals. The dataset from normal subjects are recorded for two active cognitive tasks during each recording session. 1-Meditation activity: The subject is asked to meditate for a fixed period of time while his brain waves are recorded. 2-Math activity: The subject is given non-trivial multiplication problems, such as 79 times 56 and is asked to solve them without vocalizing or making any other physical movements. The problems were designed so that they could not be solved in the time allowed.
- 21. The EEG data is segmented. Channel spectral power for three spectral bands Alpha, Beta and Gamma is computed. 14 x 3 = 42 features are extracted for each segment of the data. Using PCA and PSD. The unit of PSD is energy per frequency (width). Computation of PSD can be done directly by the method of Fourier analysis or computing autocorrelation function and then transforming it.
- 22. The Discrete Fourier transform is given by where (f1, f2) is the frequency band and Sx(f) is the power spectral density. The inter-hemispheric channel spectral power differences in each spectral band are given by P 𝑑𝑖𝑓𝑓 = (P1 – P2) / (P1 + P2) where P1 and P2 are the powers in different channels in the same spectral band but in the opposite hemispheres.
- 23. The obtained feature vector is compared against a previously stored feature vector for that subject, using Euclidean Distance for template matching. The match is considered good if the result of the comparison >0.78 keeping in mind the need to satisfy low False Acceptance Error (FAE) and False Rejection Error (FRE).
- 24. The authentication system was realized by developing an application which would lock and unlock the screen. Initially the screen is locked and a subject’s EEG signals for two mental tasks are recorded and stored as a reference, called the training phase. If the screen is to be unlocked, the subject’s brain waves are recorded again and matched with the earlier stored sample. If there is a considerable match, then the screen is unlocked, otherwise it will stay locked.
- 25. The description of the working prototype is outlined as: 1-Training of the system 2-Feature extraction 3-Creating user profile 4-Authenticating The feature extraction and matching part are coded in MATLAB, while the UI part is designed and coded in C#.
- 27. Steps Step 1: The initial screen which is the main prompt screen facilitates the user to perform the lock screen, add/remove user, change account name and restore activities. Step 2: We add a new user as there are no existing users initially. The training form opens wherein we train the system for authentication. The training is based on two activities, Meditation and Math activity. While the subject is performing these activities, the signals are recorded and stored. Main Prompt Window.
- 28. Step 3: Once the training process is complete, the user returns to the main prompt form .The user can now lock the screen by clicking on the lock screen option. The login form appears wherein user name has to be specified for unlocking the screen. There are 3 available options, Unlock, Restart and Shutdown. User login window
- 29. Step 4: When the unlock option is pressed by the user an authentication form appears. Two activities, for which the system has been trained earlier, must be performed for authentication , one after the other. Step 5: If the authentication is successful then the main prompt form is displayed and the screen is successfully unlocked, else the authentication fails and the screen remains in the locked state.
- 30. Abstract 1-INTRODUCTION 2-RELATED WORK 3-METHODOLOGY 3-1 Data Acquisition 3-2 Preprocessing and Feature Extraction 3-3 Classification 3-4 Implementation 4-CONCLUSION
- 31. the EEG can be used for biometric authentication. Person authentication aims to accept or reject a person claiming an identity. We perform EEG recording, feature extraction and matching of the feature vector with the stored feature vector, all in real time. This system is designed without using any type of external stimulus. This work, however, needs more refinement such as, i. Recording must be done in clinical conditions where there are no external interferences (noise free environment). ii. Training the users to perform the various mental tasks with full concentration. iii. Handling high dimensional data. iv. Devising a more or less perfect matching algorithm that gives 0 FAE and 0 FRE.