PRISM :: Browsing by Author "Gavrilova, Marina"

Browsing by Author "Gavrilova, Marina"

Now showing 1 - 20 of 33

Open Access
A MULTIMODAL BIOMETRIC SYSTEM BASED ON RANK LEVEL FUSION
(2013-01-07) MONWAR, MD. MARUF; Gavrilova, Marina
In recent years, biometric based security systems achieved more attention due to continuous terrorism threats around the world. However, a security system comprised of a single form of biometric information cannot fulfill users’ expectations and may suffer from noisy sensor data, intra and inter class variations and continuous spoof attacks. To overcome some of these problems, multimodal biometric aims at increasing the reliability of biometric systems through utilizing more than one biometric in decision-making process. In order to take full advantage of the multimodal approaches, an effective fusion scheme is necessary for combining information from various sources. Such information can be integrated at several distinct levels, such as sensor level, feature level, match score level, rank level and decision level. In this doctoral research, I present a new methodology based on fusion at the rank level, which is a relatively new approach compared to others, to combine multimodal biometric information from three biometric identifiers (face, ear and iris). I investigate different rank fusion methods, such as highest rank, Borda count and logistic regression. I introduce a novel rank fusion algorithm based on Markov chain which significantly increases the recognition performance of the multimodal biometric system, can handle partial ranking lists, and satisfies the Condorcet criteria essential for fair ranking process. In order to increase the processing speed and to obtain the level of confidence of recognition outcomes of the multimodal biometric system, I further employ fuzzy logic based fusion for biometric authentication. The fuzzy fusion method is based on fuzzy logic and uses match score and rank information of the multimodal biometric system. The experiment results tested within different multimodal biometric database framework show superiority of the proposed approaches to other biometric information fusion methods. The developed system can be effectively used by security and intelligence services for controlling access to prohibited areas and protecting important national or public information.
Open Access
AestheticID: Human Identification Using Audio-Visual Preferences
(2024-11-14) Iffath, Fariha; Gavrilova, Marina; Sousa, Mario Costa; Tepperman, Charles
Over the last decade, Online Social Media platforms have witnessed a substantial expansion due to the extensive reliance of individuals on these communication channels. These platforms are widely utilized to convey emotions, share opinions, and express preferences through various means such as artworks, multimedia content, and blogs. These individual-specific traits have a wide range of applications such as personalized recommender systems, human behavior analysis, human-computer interaction, robotics, and biometric security. Aesthetic biometric systems utilize users’ unique preferences towards various subjective forms such as images, music, and textual content. This study introduces a novel deep learning-based multi-modal aesthetic system, with a primary contribution to the development of an attention-based fusion method for person identification. The proposed identification system leverages a deep pre-trained model for high-level feature extraction from visual and auditory modalities. The paper introduces a novel fusion architecture named attention-based residual fusion network (ARF-Net) to incorporate two heterogeneous aesthetic modalities. The proposed system is validated on two proprietary aesthetic datasets outperforming the existing state-of-the-art aesthetic biometric systems for person identification. The proposed architecture stands out for its efficiency, showcasing a lightweight architecture with minimal parameters, ensuring optimal performance across multiple aesthetic modalities.
Open Access
Analysis of a Vertical-Axis Tidal Turbine Using the Variational Multiscale Formulation
(2022-09) Dhalwala, Musaddik; Korobenko, Artem; Gavrilova, Marina; Wood, David
This thesis first investigates the performance and near-wake characteristics of a full-scale vertical-axis tidal turbine under a uniform inflow and turbulent inflow with a 5% and 10% turbulence intensity. The governing equations of the flow field are the incompressible Navier-Stokes equations. As the turbine rotates throughout the simulation, these equations are expressed in a slightly different form referred to as the arbitrary Lagrangian-Eulerian (ALE) framework. The purpose of the ALE framework is to allow the mesh to move arbitrarily while the fluid moves independently of the mesh motion. From the available large eddy simulation (LES) formulations in the literature, the variational multiscale (VMS) formulation is used to discretize the system of equations. Unlike classical LES, the VMS formulation does not have any problems with specifying an appropriate filter for different flows. To study the effect of a turbulent inflow, a turbulence generation method referred to as Smirnov's random flow generation (RFG) is used. From the numerous turbulence generation methods available, Smirnov’s RFG was chosen as it can generate a turbulent velocity field that is divergence-free. A divergence-free velocity field ensures compatibility with the incompressible Navier-Stokes equations that govern the flow field and results in good numerical stability. While the performance of the turbine slightly reduced under a turbulent inflow compared to a uniform inflow, there was a negligible difference in its performance between the two turbulent inflow conditions. A turbulent inflow also resulted in large fluctuations of the instantaneous power coefficient. Lastly, the wake recovery was notably improved under a turbulent inflow. Next, the effect of a free surface on the performance and flow field of the turbine with different blade-strut configurations is studied. There was a negligible effect of the free surface on turbine performance and the flow field during deep immersion. Moreover, the tip-struts configuration was 15% more efficient than the quarter-struts configuration under deep immersion. Under shallow immersion, the performance of both blade-strut configurations reduced.
Open Access
Analysis of Deep Domain Adaptation Methods for Brain Magnetic Resonance Image Segmentation
(2022-12-16) Saat, Parisa; Hemmati, Hadi; Souza, Roberto; Gavrilova, Marina; Deshpande, Gouri
Accurate brain segmentation is critical for magnetic resonance imaging (MRI) analysis pipelines. Machine-learning-based brain MR image segmentation methods are among the state-of-the-art techniques for this task. Nevertheless, the segmentations produced by machine learning models often degrade in the presence of expected domain shifts between the test and train sets data distributions. These domain shifts are expected due to several factors, such as scanner hardware and software differences, technology updates, MRI acquisition parameters, and differences across the scanned populations. Domain adaptation (DA) methods can make machine learning models more resilient to these domain shifts. In this thesis, I investigated supervised and unsupervised deep domain adaptation methods for brain MRI segmentation. Two scenarios are investigated. In the first scenario, data shifts occur due to hardware and software differences across different MRI scanner vendors (General Electric, Philips, and Siemens). In the second scenario, data shifts occur due to differences in the scanned populations. The source brain MRI data comes from adults, while the target data corresponds to pediatric patients, whose brains are still developing. The main findings of this thesis are that there is not a single DA technique that consistently outperforms others, and hyperparameter tuning and computational times for these methods still need to be addressed before adopting these methods in clinical practice. Another important finding is that the DA techniques worked better for data shifts resulting from hardware and software differences across different MR scanner vendors than data shifts from population differences. The labeled data and source code used in this thesis were made publicly available and serve as a benchmark for evaluating DA methods for brain MRI segmentation.
Open Access
Bi-Modal Deep Neural Network for Gait Emotion Recognition
(2022-11-23) Bhatia, Yajurv; Gavrilova, Marina; Jacobson, Michael; Runions, Adam
Emotion Recognition systems can be used for autonomous tasks such as video gaming experiences, medical diagnosis, adaptive education, and smart homes. Several biometric modalities, including face, hands, and voice have been successfully used for emotion recognition tasks. Gait Emotion Recognition (GER) is an emerging domain of research that is focused on identifying the emotional state of a person from gait biometric, which represents the person’s manner of walking. In comparison to the other modalities, gait provides a non-intrusive method to collect data remotely without an expert’s supervision. Moreover, unlike facial expression-based emotion recognition, it does not require high-resolution data for inference. Early works in GER produced limited feature sets and used classical machine learning methodologies to infer emotions, but could not achieve high performance. This thesis proposes powerful architectures based on deep-learning to accurately identify emotions from human gaits. The proposed Bi-Modal Deep Neural Network (BMDNN) architecture utilizes robust handcrafted features that are independent of dataset size and data distribution. The network is based on Long Short-Term Memory units and Multi-Layered Perceptrons to sequentially process raw gait sequences and facilitate feature fusion with the handcrafted features. Lastly, the proposed Bi-Modular Sequential Neural Network (BMSNN) has a low number of parameters and a low inference time, hence making it suitable for deployment in real world applications. The proposed methodologies were evaluated on the Edinburgh Locomotive MoCap Dataset and outperformed all recent state-of-the-art methods.
Open Access
Chaotic Neural Network for Biometric Pattern Recognition
(2012-08-30) Ahmadian, Kushan; Gavrilova, Marina
Biometric pattern recognition emerged as one of the predominant research directions in modern security systems. It plays a crucial role in authentication of both real-world and virtual reality entities to allow system to make an informed decision on granting access privileges or providing specialized services. The major issues tackled by the researchers are arising from the ever-growing demands on precision and performance of security systems and at the same time increasing complexity of data and/or behavioral patterns to be recognized. In this paper, we propose to deal with both issues by introducing the new approach to biometric pattern recognition, based on chaotic neural network (CNN). The proposed method allows learning the complex data patterns easily while concentrating on the most important for correct authentication features and employs a unique method to train different classifiers based on each feature set. The aggregation result depicts the final decision over the recognized identity. In order to train accurate set of classifiers, the subspace clustering method has been used to overcome the problem of high dimensionality of the feature space. The experimental results show the superior performance of the proposed method.
Open Access
Confidence-Based Rank Level Fusion For Multimodal Biometric Systems
(2015-11-17) Talebi, Hossein; Gavrilova, Marina
In recent years, the inevitable need for reliable biometric identity management systems in applications such as border crossing, welfare distribution, and accessing critical facilities has drawn researchers' attention to the area of biometric. The intrinsic limitations of unimodal biometric systems such as non-universality, sensitivity to noisy sensor data, inter and intra class variations and spoof attacks have resulted in significant attention toward multimodal biometric systems. An important aspect of a multimodal biometric system is the fusion of information from multiple biometric sources. This thesis focuses on using the notion of Resemblance Probability Distributions to calculate confidence measures for different biometric matchers and use these confidence measures in the fusion module to improve the identification rate of the system. This thesis approaches the problem of low inter class variation and low quality data by proposing Rank List Reinforcement and Confidence-based Ranked List Selection methods.
Open Access
Context-based gait recognition
(2013-03-06) Bazazian, Shermin; Gavrilova, Marina
With the increasing demand for automatic security systems capable of recognizing people from a far distance and with as less cooperation as possible, gait recognition emerged as a very popular behavioral biometric because it is remotely observable and unobtrusive. However, the complexity and the high variability of gait patterns limit the power of gait recognition algorithms and adversely affect their recognition rates. Aiming to improve the performance of gait recognition systems without sacrificing the main advantages of gait, in this thesis, I introduce a novel multimodal gait recognition system that combines the gait patterns of the subjects with the context data related to their behavioral and social patterns. To the best of my knowledge, this is one of the only examples that the social patterns of the subjects have been used as a source of information in a multimodal biometric system. This thesis introduces a well-defined framework for defining, modeling, learning, storing and matching context data in a gait recognition system. The proposed behavioral modeling and matching framework is very flexible and can easily be adapted to different applications and multimodal biometric systems. According to the conducted experiments, the proposed gait recognition system can achieve significant improvements in the performance at a very low computational cost. The comparison of the method with other existing methods in the same area shows that the proposed approach is applicable and effective.
Open Access
Decoding Identity Through Text-based Human Micro-expressions: A Novel Approach in Social Behavioral Biometrics
(2023-06) Wahid, Zaman; Gavrilova, Marina; Farhad, Maleki; John Jacobson Jr., Michael
In recent years, Social Behavioral Biometrics (SBB) has gained prominence due to the dramatic changes in the way people socialize in this technologically-advanced era. The reliance on Online Social Networks (OSN) for formal and informal social interactions has become the norm. This thesis introduces a novel SBB trait, human micro-expression, for online person identification. An emotion detection model is initially developed to extract Parrott’s primary emotion scores from OSN users’ writing samples posted on Twitter. The corresponding emotion-progression features are extracted using an original technique that turns users’ microblogs into emotion signals. The Dynamic Time Warping (DTW) algorithm is utilized to facilitate the process of emotion-progression feature extraction across OSN users’ emotion signals trajectories. Then, a unimodal SBB system based on the proposed human micro-expression biometric is implemented, leveraging rank-level weighted Borda count to improve the performance of person identification. Furthermore, a multimodal SBB system is proposed that incorporates the proposed SBB trait into original SBB traits in state-of-the-art. To evaluate the effectiveness of the proposed system, a proprietary benchmark dataset consisting of 250 Twitter users is employed. The experimental results demonstrate that the novel human micro-expression trait exhibits strong distinguishability among OSN users and can be used for person identification. Moreover, the study reveals that the proposed social behavioral biometric outperforms the majority of original SBB traits, indicating its potential value in future research and applications. The proposed multimodal SBB system exhibits superior performance compared to existing state-of-the-art multimodal SBB systems, further emphasizing the utility of incorporating the human micro-expression trait in multimodal SBB systems to improve person identification performance. This thesis contributes to the burgeoning field of social behavioral biometrics, with the potential for significant advancements in future research on person identification and online security.
Open Access
DeepCADe: A Deep Learning Architecture for the Detection of Lung Nodules in CT Scans
(2018-01-16) Golan, Rotem; Jacob, Christian; Denzinger, Joerg; Gavrilova, Marina; Frayne, Richard; Cunningham, Ian
Early detection of lung nodules in thoracic Computed Tomography (CT) scans is of great importance for the successful diagnosis and treatment of lung cancer. Due to improvements in screening technologies, and an increased demand for their use, radiologists are required to analyze an ever increasing amount of image data, which can affect the quality of their diagnoses. Computer-Aided Detection (CADe) systems are designed to assist radiologists in this endeavor. In this thesis, we present DeepCADe, a novel CADe system for the detection of lung nodules in thoracic CT scans which produces improved results compared to the state-of-the-art in this field of research. CT scans are grayscale images, so the terms scans and images are used interchangeably in this work. DeepCADe was trained with the publicly available Lung Image Database Consortium and Image Database Resource Initiative (LIDC-IDRI) database, which contains 1018 thoracic CT scans with nodules of different shape and size, and is built on a Deep Convolutional Neural Network (DCNN), which is trained using the backpropagation algorithm to extract volumetric features from the input data and detect lung nodules in sub-volumes of CT images. Considering only lung nodules that have been annotated by at least three radiologists, DeepCADe achieves a 2.1% improvement in sensitivity (true positive rate) over the best result in the current published scientific literature, assuming an equal number of false positives (FPs) per scan. More specifically, it achieves a sensitivity of 89.6% with 4 FPs per scan, or a sensitivity of 92.8% with 10 FPs per scan. Furthermore, DeepCADe is validated on a larger number of lung nodules compared to other studies (Table 5.2). This increases the variation in the appearance of nodules and therefore makes their detection by a CADe system more challenging. We study the application of Deep Convolutional Neural Networks (DCNNs) for the detection of lung nodules in thoracic CT scans. We explore some of the meta parameters that affect the performance of such models, which include: 1. the network architecture, i.e. its structure in terms of convolution layers, fully-connected layers, pooling layers, and activation functions, 2. the receptive field of the network, which defines the dimensions of its input, i.e. how much of the CT scan is processed by the network in a single forward pass, 3. a threshold value, which affects the sliding window algorithm with which the network is used to detect nodules in complete CT scans, and 4. the agreement level, which is used to interpret the independent nodule annotations of four experienced radiologists. Finally, we visualize the shape and location of annotated lung nodules and compare them to the output of DeepCADe. This demonstrates the compactness and flexibility in shape of the nodule predictions made by our proposed CADe system. In addition to the 5-fold cross validation results presented in this thesis, these visual results support the applicability of our proposed CADe system in real-world medical practice.
Open Access
Efficient Image Matching using Regions of Interest
(2014-09-12) Bhattacharya, Priyadarshi; Gavrilova, Marina
This thesis tackles the challenging problem of producing a ranked list of images, based on similarity to a query image in a large, unordered image collection. The application domain considered spans from landmarks and scenes to general objects. Existing state-of-the-art methodology for object retrieval in large image collections [SZ03] [PCI+07], based on the bag-of-words approach, has its limitations. Discarding spatial information about features in the image representation stage results in false matches. Spatial verification, used as a post-processing step to improve retrieval accuracy, is computationally expensive. Being based on a global model of the image, the method is susceptible to noise and background clutter. In this thesis, I propose a novel image modelling methodology, that is driven by attention to interesting regions of an image and representing these regions at a high level of detail. Rich spatial information about features is injected in the image modelling stage. This eliminates the need for computationally expensive spatial verification as a post-processing step. A novel image matching methodology is proposed that matches localized regions in images, instead of matching images at a global level using a histogram-based approach. The motivation is that, despite large changes in their global appearance, some of the regions in the images will still match well. The proposed methodology is observed to be highly robust to viewpoint changes, occlusion and background clutter and suitable for sub-image retrieval. It allows real-time search and is scalable to large image corpuses. An added advantage is that object localization is possible simultaneously with search, with minimal computing effort. Experiments reveal the superior performance of the proposed methodology over state-of-the-art methods that utilize spatial information. It is also several orders of magnitude faster than the bag-of-words approach with spatial verification.
Open Access
Gauss–Laguerre wavelet textural feature fusion with geometrical information for facial expression identification
(Spinger Open, 2012-09-25) Poursaberi, Ahmad; Noubari, Hossein Ahmadi; Gavrilova, Marina; Yanushkevich, Svetlana N
Open Access
Icosahedral Maps for a Multiresolution Representation of Earth Data
(2016) Jubair, Mohammad Imrul; Alim, Usman; Gavrilova, Marina; Pidlisecky, Adam
The icosahedral non-hydrostatic (ICON) model is a climate model based on an icosahedral representation of the Earth and is used for numerical weather prediction. In this thesis, we investigate the unstructured representation of different cells in ICON and undertake the task of designing a technique that converts it to a common structured representation. We introduce icosahedral maps, data structures that are designed to fit the geometry of cells in the ICON model irrespective of their types. These maps represent the connectivity information in ICON in a highly structured two-dimensional hexagonal representation that provides explicit neighborhood information. Our maps facilitate the execution of a multiresolution analysis on the ICON model. We demonstrate this by applying a hexagonal version of the discrete wavelet transform in conjunction with our icosahedral maps to decompose ICON data to different levels of detail and to compress it via a thresholding of the wavelet coefficients.
Open Access
Integrating Multi-Domain Electronic Health Data, Machine Learning, and Automated Cardiac Phenomics for Personalized Cardiovascular Care
(2024-04-19) Dykstra, Steven; White, James; Gavrilova, Marina; Wilton, Stephen; Alim, Usman
This thesis aims to address core challenges surrounding the integration of multi-domain cardiovascular data, inclusive of patient reported health, electronic health information, and diagnostic imaging, to support artificial intelligence (AI) based risk prediction modelling. Despite inaugural success surrounding the use of AI-driven approaches to leverage granular features from each respective data source, the lack of integration continues to limit a comprehensive representation of patient health critical to the implementation of AI-augmented clinical decision support (AI-CDS). Central to this thesis was the primary hypothesis that patient-consented migration, integration, and curation of disparate data sources can be achieved in real-world clinical environments, permitting longitudinal accumulation of standardized resources for machine learning-based risk modelling. To test this hypothesis, my first aim was to develop a software infrastructure to establish and maintain a precision health data model for cardiovascular care. This data model forms the foundation of the Cardiovascular Imaging Registry of Calgary (CIROC), a platform which to date has generated structured data resources for over 20,000 unique patients with cardiovascular disease. The success of this robust data model has led to the expansion of this infrastructure to support all clinics of the Libin Cardiovascular Institute. The design of this initiative, called the PULSE program, was established as an objective of Aim 1, delivering a structured manuscript describing methods and recommendations for implementing a scalable institutional personalized medicine program for the ethical, fair, and equitable introduction of AI-CDS. Subsequently, the second aim demonstrates the value of the established data model, highlighting how it can be used for the development and validation of machine-learning based prediction models for cardiovascular outcomes. Utilizing multi-domain features of the CIROC data model, I demonstrated superiority of machine learning-based approaches over traditional risk prediction methods to predict new-onset atrial fibrillation, a leading cause of stroke. This study highlighted the value of integrating patient-reported health, electronic health record, and cardiac diagnostic data to forecast future cardiovascular events with improved accuracy. Further, my third aim targeted an expansion of disease features from source diagnostic testing data to improve risk modelling. To achieve this, I developed deep learning-based models for the automated analysis (segmentation and fiducial labelling) of the left ventricle from cine cardiac MRI imaging, enabling the delivery of 3D shape phenomics. This work showcases the capacity for deep learning techniques to further enhance the developed data models for patient-specific risk modelling by supporting advanced analyses of unique disease characteristics including shape and deformation. This novel solution is now planned for external validation by a large, international clinical study assessing the incremental value of 3D shape phenomics to improve prediction accuracy across a broad range of diseases. Overall, this thesis presents a comprehensive exploration of technical development required for, and value generated by multi-domain data integration for AI-CDS in cardiovascular care. Incremental to demonstrating feasibility, the deliverables of this thesis serve as a foundation for growth of an emerging institutional precision medicine initiative and for the development of future advanced multi-domain machine learning models relevant to cardiovascular care.
Open Access
Iris Synthesis: A Reverse Subdivision Application
(2005-09-26) Wecker, Lakin; Samavati, Faramarz; Gavrilova, Marina
Due to renewed interest in security, iris images have become a popular biometric alternative to fingerprints for human identification. However, there exist very few databases on which researchers can test iris recognition technology. We present a novel method to augment existing databases through iris image synthesis. A multiresolution technique known as reverse subdivision is used to capture the necessary characteristics from existing irises, which are then combined to form a new iris image. In order to improve the results, a set of heuristics to classify iris images is proposed. We analyze the performance of these heuristics and provide preliminary results of the iris synthesis method.
Open Access
MARIS - Capabilities and Applications for Security and Risk Analysis
(2006-02-15) Jensen, April; Gavrilova, Marina
This report contains a description of goals accomplished over the course of the project. It also contains descriptions of ideas for data analysis and progress made regarding these goals. New data that was created is described, including how it was generated, where it is stored and some particular goals that it may prove useful for. Finally, some figures are included to illustrate some of the uses of both ArcGIS and MARIS for the display and analysis of data with the goals of risk analysis and security.
Open Access
Model-Based Gait and Action Recognition Using Kinect
(2016) Ahmed, Faisal; Gavrilova, Marina; Gavrilova, Marina; Alim, Usman; Mintchev, Martin
Being the very first in the category of low-cost consumer-level depth sensors, the recent release of Microsoft Kinect has opened the door to a new generation of computer vision and biometric security applications. This thesis focuses on designing new methodologies for Kinect-based gait and action recognition systems that utilize the Kinect 3D virtual skeleton to construct effective and robust motion representations. The proposed gait recognition method focuses on designing a feature descriptor that can capture person-specific distinct motion patterns, caused by the influence of human physiology and behavioral traits. On the other hand, the proposed action recognition method involves constructing a person-independent feature descriptor that can suppress person-specific motion traits while highlighting a more generic and high level description of action-specific skeletal joint movements. Extensive experiments with three recently released public benchmark databases demonstrate the effectiveness of the proposed methodologies, compared against state-of-the-art gait and action recognition methods.
Open Access
Multidimensional Projection Visualization: Control-points Selection and Inverse Projection Exploration
(2016) Portes dos Santos Amorim, Elisa; Costa Sousa, Mário; Samavati, Faramarz; Gavrilova, Marina; Jacob, Christian J.; Rios, Cristian; Esperanca, Claudio
The task of interpreting multidimensional data is as important as it is challenging. The importance comes from the fact that virtually every data worth analyzing is multidimensional, while the challenge comes from the very nature of these data sets, as the multiple features describing each instance can quickly overwhelm our visual perception system, thus making it difficult to observe meaningful information. Visualization techniques play an essential role in simplifying this task, by preprocessing the data to extract critical features and displaying them effectively, by using visual metaphors that can be easily understood. Multidimensional Projection (MP) is one of such techniques, whose fundamental goal is to present an overview of the data distribution in the form of a 2D scatterplot graph. It does so by reducing the dimensionality of the dataset in such a way that distances are preserved as much as possible. MP approaches, along with most visualizations, are shifting from a static display to a more interactive one, allowing human intervention to modify the layout and facilitate exploration and understanding of the data. In this thesis, I present contributions that specifically relate to interactive aspects of multidimensional projection. First, I propose a computational framework and methodology for control points selection. Control points are a particular set of projected points used to steer and rearrange the projection layout. I demonstrate the proposed method can improve the projection quality while requiring only a small amount of control points. Second, I introduce inverse projection, a novel paradigm to create multidimensional points exclusively through 2D interactions. The projection space is transformed into a canvas, where new points can be added. These new points are then mapped into the original multidimensional space, i.e., they become unique multidimensional instances themselves. Lastly, I present the usability of the inverse projection framework in two demonstration examples. (1) A parameter exploration prototype system for optimization with multiple minima. (2) A face-synthesis application, where new face models are generated on the fly.
Open Access
Multimodal Cancelable Biometric System
(2016) Paul, Padma Polash; Gavrilova, Marina; Elhajj, Reda; Wang, Yingxu; Yanushkevich, Svetlana; Howard, Newton
Cyberattacks against individuals and organizations are increasing at alarming rates all over the world. Traditional password, pin, smart id, and tokens based systems are insufficient to provide reliable authentication. Biometric authentication is now widely used in both physical and virtual worlds. Unfortunately, even the well-established biometric systems are suffering from vulnerabilities, with the most crucial components being biometric templates that store user biometric data. If the biometric of a user is compromised, the identity and privacy of an individual are compromised as well, since it is impossible to revoke or reissue the biometric template. Therefore, secure and revocable biometric template generation algorithms are required to ensure biometric system integrity and user privacy. In this thesis, biometric template generation algorithm for multimodal biometric system is presented. To achieve the template protection for the multimodal biometric verification system, new methods for feature level fusion and feature extraction are proposed, called Cancelable Feature Fusion (CFF) and Cancelable Binary Pattern (CBP), respectively. CFF combines multiple biometric traits using random indexes so that for every fusion it generates a new fused template. Developed 2-Fold Cross-Folding (2-CF) and Generalized Cross-Folding (G-CF) are new algorithms for cancelable feature fusion, which utilize random indexes to combine multiple biometric traits. Another developed method is CBP, a biometric feature extraction algorithm that can generate a new set of features from a single sample as needed. For after matching fusion, Social Network Analysis (SNA) based score fusion is proposed to achieve better verification accuracy. In this thesis, traditional feature fusion and feature extraction algorithms are replaced with CFF and CBP respectively to support the template protection. For the validation of the methodologies, genuine, stolen and fake key scenarios are analyzed using several sets of virtual multimodal biometric databases of face, ear, and signature. Experimental results show that proposed Multimodal Cancelable Biometric System (MCBioS) architectures can achieve 0% Equal Error Rate (EER) and a higher template integrity.
Open Access
Normal vs. Abnormal Behavior
(2006-02-15) Jensen, April; Gavrilova, Marina
The term outlier refers to data that deviates significantly from what is considered normal. Outliers are often of interest since they may indicate that data was generated by a mechanism other than the expected one [23], or may indicate abnormal or suspicious behavior. The modeling and analysis of normal vs. abnormal behavior proves useful in security-related areas such as biometrics, video surveillance and intrusion-detection, as well as in other areas such as medical imaging, among others.