Biometric System for Person Identification: Exploring Unimodal and Multimodal Techniques

Abstract

Biometric representation of humans often require the tasks such as identification and verification using human computer interaction (HCI), which can be achieved using various modalities such as fingerprint, face, retina, voice, signature, etc. However, multiple attacks can challenge the security and performance of existing biometric systems. Electroencephalography (EEG) is considered an alternative to developing a robust biometric system. Brain activities represented using EEG signals are more sensitive, secure, and difficult to copy and steal. We have studied various aspects and understood the shortcomings in the existing schemes. Many person identification systems have been designed using machine learning and deep learning approaches in the existing literature. The available biometric systems need performance improvement, keeping security into consideration. In this research, we have tried to address these issues towards improving the system’s overall performance. In this thesis, a Spatio-temporal dense architecture for EEG-based person identification/authentication is designed and explored. In the proposed scheme, the raw EEG data are processed to extract robust and informative spatial features using Convolutional Neural Networks (CNN), known for automatic feature extraction from the raw data. Then, a long short-term memory (LSTM) network is utilized to process temporal data, and person identification is carried out. The experiment has been carried out on a publicly available dataset consisting of an EEG of 109 subjects. The architecture is tested on two baseline situations, i.e., eye closed (EC) and eye opened (EO). A Spatio-temporal identification/ authentication model for a publicly available emotion-based DEAP (Database for Emotion Analysis;Using Physiological Signals) dataset has been designed to explore and prove the robustness. Person identification rates of 99.95%, 98%, and 93.87% have been recorded for EC, EO states, and emotion using the proposed scheme. Experimental results demonstrate the robustness of the proposed scheme in terms of person identification and outstrip existing works. EEG-based biometric are putting forward solutions because of their high-safety capabilities and handy transportable instruments. In the Spatio-temporal model, more parameters and dimensionality constraints affect the overall training performance. Motor imagery EEG (MI-EEG) is a broadly centered EEG signal exhibiting a subject’s motion intentions without actual actions. This research proposes an unsupervised framework for feature learning based on autoencoder. It learns sparse feature representations for EEG-based person identification. Autoencoder-CNN exhibits the person identification task for signal reconstruction and recognition. The framework proved to be a practical approach in managing the massive volume of EEG data and identifying the person based on their different task with resting states. The experiments have been conducted on the standard publicly available Motor imagery EEG dataset with 109 subjects and an emotion-based dataset with 32 users. Different task and emotion-based models using autoencoder and CNN have been explored in this research. We have noted the highest recognition rate of 87.60% for task based identification, 98.43% for the emotion-based identification model. Moreover, a maximum 99.89% recognition rate for resting-state has been recorded using the Autoencoder-CNN model. The outcomes imply that the overall performance of the proposed framework is similar or advanced to that of a novel method. The shape is proved to be a realistic technique to control the full-size extent of EEG data and pick out the individual, primarily based on their specific task. In the above-proposed schemes, a single trait is used for designing person identification systems. This can be made much more secure and reliable using more traits, which is addressed in this research to explore multimodality. A novel multimodal biometric person identification system by using two closely connected traits, i.e., signature and brain signals as Electroencephalography (EEG), is proposed in this research. The strokes while signing stimulates the brain signals. The response of the activity is unique for a user in the brain, and this relationship is explored in this research in designing a more secure and robust person identification system. The performance accuracy of 97.54% for CNN, 98.61% for CNN-LSTM, and 98.74% for CNN with PCA are observed. To improve the recognition rate and to secure the authentication approaches, multimodal biometric is proved as a novel technique.