Security technologies that measure our physical attributes are called biometrics

3.1 Behavioral Biometrics

Biometrics involves the study of approaches and algorithms for uniquely recognizing humans based on physical or behavioral cues. Traditional approaches are based on fingerprint, face, iris, and can be classified as physiological biometrics, that is, they rely on physical attributes for recognition. These methods require cooperation from the subject for collection of the biometric. Recently, “behavioral biometrics” have been gaining popularity, where the premise is that behavior is as useful a cue to recognize humans as their physical attributes. The advantage of this approach is that subject cooperation is not necessary and it can proceed without interrupting or interfering with the subject's activity. Since observing behavior implies longer term observation of the subject, approaches for action recognition extend naturally to this task. Currently, the most promising example of vision-based behavioral biometric is human gait [7].

2.6.2.2 Biometric cryptosystems

Biometric cryptosystems associate a cryptographic key with the biometric data presented during the enrollment process, which simplifies aspects of a biometric system with regards to security. Due to the variability of biometric data, direct key extraction is rather difficult and henceforth a system utilizing a biometric cryptosystem allows the existence of partial data of enrolled biometrics (enough not to compromise the system); this is termed helper data and it assists in key extraction during the authentication process. There are a few instances of deep learning methods employed in biometric cryptosystems. The model discussed for facial recognition earlier utilizes concepts of biometric cryptosystems. Xiulai Li proposed a deep learning-based method [71], where the iris dataset was normalized and used to train CNN architecture-based deep learning neural network models. The encryption side collected the iris image and inputed the trained deep learning model to extract the feature vector that was passed through the Reed–Solomon encoder that encoded the given feature vector. Albakri and others [73] proposed a CNN-based biometric cryptosystem for the protection of a blockchain's private key. Here, the biometric input of users other than the training input was taken, its features were extracted, and a vault was created to mix these features along with a key for protection.

6.1.1 Introduction

Biometrics are measurable physiological or behavioural characteristics that can be used to verify the identity of an individual. There are a wide variety of different types of biometric available for measurement. The most popular include a person’s fingerprint, face, hand, iris, voice, signature, retina or typing rhythm. However other biometrics are under research, such as gait recognition (the way a person walks), earprint analysis, DNA recognition or even recognising a person from their spectral characteristics (which is a measure of the composition and structure of a person’s skin and underlying tissue).

The biometric industry produces systems that use a person’s biometric, or a combination of biometrics, to automatically check that person’s identity.

3.5.7.4 Outlook

Biometrics certainly have the potential to develop well within the telecoms market place. In particular, fingerprint and speaker verification technology could prove useful forms of authentication. However, until succinct business cases – which show a definite advantage to telecoms operators – can be made, there will be moderate short-term growth within this sector. The greatest opportunities for biometrics will come in the long term – when telephone companies start taking a serious look at how to reduce fraud. By 2006 revenue is expected to be US$28 million from US$8.9 million in 2002.

1.3 Enrolment mechanisms

Biometrics are physical security mechanisms which deny any unauthorised access via authentication. This security process is referred to as biometric authentication and is reliant on individuals’ unique biological characteristics to identify the individual correctly. These traits can be used instead of passwords to verify and identify individuals as they are bound to the individual. Retina, iris, facial, fingerprint, or palm prints are all unique, but Garcia (2018) states that to preserve privacy, biometrics should never be stored in a public distributed ledger as it's accessible to anyone whether it's in an encrypted or template state. The reason for this is that there are no guarantees that the system will remain safe in future due to advances in quantum computing.

An important part of the process is the method used to compare the information presented versus the information on record as most common used methods still use signature and photo comparisons to perform most transactions. These methods are unreliable, more so when the person who is making the comparison doesn’t know the individual.

5.4 Breakdown of Companies by Geographic Region

Africa

Avalanche Technologies

Biometrics.co.za

Cyber-Card

FACE Technologies

Intervid International

Kingshad International

Asia

Arithmetic Guarding Equipments

Automa Pty

Autostar Technology

Axis Software

Banque-Tec International

BioEnable Technologies

Cecrop

ComSec Enterprises

CSIRO

Dialog Communication Systems

e-Com

Excel Systems

Fingerprint Technologies

FIST

Fusion Tech

Hectrix

Hunno Technologies

Hyundai Information Technology

IDTECK

iGuard Security Systems

ILI Technologies

Javpeetex Engineering

Keico Hi Tech Inc

Miaxis Biometrics

Nextern

NITGen

Oki Electric Industry

Optical Recognition Objectives

Print Electronics

Protocom Development Systems

Rotin Australia

Security Information Technology

Senex Technologies

Sintec Corporation

Sonda

Speech Technology Center

Startek Engineering

Suprema

Symtron Technology

Techno Imagia

ThumbAccess

Union Community

Watchvision

Yasmin Teknologi

Yuean Biometrics

Europe

A4Vision

Accu-Tech Systems

Aditech Limited

Advanced Software Technologies

AGMA

App Informatik Davos

Association for Biometrics

Astro Datensysteme

ASTRONTECH

Audata

Aurora Computer Services

Axxess Identification

Bannerbridge

Bell ID

Bell Security

Bergdata

Bioacesso

BioMet Partners

Biometri Systemer

Biometric Partners Europe

Biometric Technology Today

Biometrika

Biometrix International

Bio-Sec

BioTrust Project

BrainTrust Technology

BTG International

Cassini Advanced Systems

Cognitec Systems

Control for Doors

C-Vis

Datastrip

Dectel Security

Delsy Electronic Components

Dermalog

Digicom

Digitus

Domain Dynamics

Ensigma Technologies

EyeDentify Europe

EyeNetWatch.com

Fingerprint Cards

Florentis

Fraunhofer Institute for Integrated Circuits

Green Bit

Guardware

Heimann Biometric Systems

Idencom

IdentAlink

Idex

IDIAP

Image Metrics

Infineon Technologies

Ingenico

Integris (now Steria)

Intuate Biometrics

Keyware Technologies

Loquendo

Mason Vactron

MESA

MotionTouch

Navigator Solutions

Neuricam

Neurodynamics

Neurotechnoligija

Neusciences

Optel

Orga Card Systems

Panasonic UK

Photobase

Plettac Electronic Security

Precise Biometrics

PrintScan International

Proglobo Biometrics Technologies

Ringdale UK

ScanSoft BVBA

SD Industries

Securicor IS

Serco

Siemens Biometrics

Softpro

Software Professional

Technogama UAB

TeleTrust

Touchless Sensor Technology

TSSI

UNIDAS

Vocalis

Voicevault

Zefyr

ZN Vision Technologies

Middle East

Configate

Ideal Software

IQS Biometric Solutions

Opticom Technologies

Sentry Com

TeKey Research Group

Wondernet

North & South America

3M-AiT

Access Controls

Accu-Time Systems

AcSys Biometrics

ActivCard

Advanced Biometrics

Affinitex

AIMS Technology

ALMEX

Alternative Computer Technology

AND

Anovea Authentication Technology

Applied Biometrics Products

Atmel

Audiopoint

Aurora Biometrics

AuthenTec

Aware

Biocentric Solutions

BioconX

Bioldentix

BIO-key International

BioLink Technologies International

BioLynx

Biometric Access

Biometric Associates

Biometric Consortium

Biometric Digest

Biometrica Systems

Biometrics Direct

Biometrics Imagineering

BioNetrix System

BioPay

Bioscrypt

BioThentica

Business Systems Engineering

Cansec Systems

Cavio

Cherry

Clinisync

Cogent Systems

Communication Intelligence

ComnetiX Computer Systems

CompuBlox

Control Module

Cross Match Technologies

Cyber-SIGN

Daon

Data Management

Datacard Group

Datakey Electronics

DataTreasury

DBA Systems

DelSecur

Digimarc ID Systems

Digital Descriptor Systems

Digital Persona

East Shore Technologies

ECryp

Ethentica

Evive

Ex Cle

Exact Identification

exResource

EyeTicket

FaceBase

Fidelica Microsystems

Financial Information Systems

FingerPrint USA

FingerSec

Geometrix

GEZ Microsystems

Grapho Technologies

Griaule Fingerprint Recognition

I/O Software

IBIA

IBMC

ID Technologies

IDentification Systems

Identix

IDynta Systems

ImageWare Systems

Imaging Automation

Imagis Technologies

Inception Technologies

Indivos

Inroad Solutions

Interactive Products

International Biometrics Group

International Electronics

InterVoice-Brite

IQ Biometrix

Iridian Technologies

J. Markowitz Consultants

KeyTronicEMS

Kinetic Sciences

Konetix

Labcal Technologies

Lockheed Martin Information

Technology

Lone Wolf Software

Lumidigm

Microsoft

Motorola Semiconductor Products

National Biometric Test Center

NCT Group

NEC Technologies

Net Nanny Software International

Nexus

NIST

Novell (NMAS)

NOVUS Technology

Nuance Communications

Omron Technology Ventures Group

Ottawa Telephony Group

Oxford Micro Devices (OMDI)

Persay

Polaroid ID

Printrak International

Prism Consulting

Privacy Curtain

Qvoice

Ravco Security

Recognition Systems

Retinal Technologies

SAFLINK

Sagem Morpho

SecuGen

SecureTech Solutions

Security Biometrics

Sense Holdings

Sentry KIDS FingerTIPS

Siemens PSE TecLab

[email protected]

Simple Technology

Smart Biometrics

Sonetech

SpeechWorks

SQN Banking Systems

STMicroelectronics

Stromberg

SyntheSys Secure Technologies

TASC

Tennyson Biometric Systems

The International Biometric Society

The Ottawa Technology Group

The Phoenix Group

Time One

TMA Associates

TMS

T-NETIX

Ultra-Scan

Unisys

Valyd

Veridicom

Veritel

VeriTouch

VeriVoice

Viisage Technology

VisionSphere Technologies

Vocent Solutions

Voice Security Systems

1 Introduction

Biometrics involves the study of approaches and algorithms for uniquely recognizing humans based on physical or behavioral cues. Traditional approaches are based on fingerprint, face, or iris and can be classified as physiological biometrics, that is, they rely on physical attributes for recognition. Physiological biometrics is usually quite accurate when acquisition is done in controlled settings. However, this turns out to be a disadvantage when one is faced with uncooperative subjects and unconstrained acquisition conditions and environments, or when one is looking for suspects in a stealthy fashion. In such cases, one requires remote acquisition techniques that may not provide sufficiently clean physiological biometrics. As an alternative, “behavioral biometrics” have been gaining popularity, where the premise is that behavior is as useful a cue to recognize humans as their physical attributes. The advantage of this approach is that subject-cooperation is not necessary and it can proceed without interrupting or interfering with the subject's activity. Observing behavior usually implies longer-term observation of the subject. In the video-based surveillance setting, this takes the form of identifying humans based on video feeds. Advances in video analysis techniques now allow one to accurately detect and track humans and faces from medium to high-resolution videos. The availability of video brings about interesting questions about how to exploit the extra information.

A number of experimental studies have shown that motion information helps in recognizing objects by enhancing the recovery of information about shape [1] or by enhancing the observer's ability to find meaningful edges [2]. It provides more views [3] and also provides information about how features change over time [4]. In the case of faces, motion has special significance, as it encodes more information than simply the 3D structure of the face. This is in the form of behavioral cues such as idiosyncratic head movements and gestures which can potentially aid in recognition tasks. Video is a rich source of information in that it can lead to potentially better representations by offering more views of the face. Further, the role of facial motion for face perception has been well documented. Psychophysical studies [5] have found evidence that when both structure and dynamics information is available, humans tend to rely more on dynamics under non-optimal viewing conditions (such as low spatial resolution, harsh illumination conditions, etc.).

Gait refers to the style of walking of an individual. Studies in psychophysics indicate that humans have the capability of recognizing people from even impoverished displays of gait, indicating the presence of identity information in gait. It is interesting, therefore, to study the utility of gait as a biometric. In fact the nature of shape changes of the silhouette of a human provides significant information about the activity performed by the human, as well as reveals idiosyncratic movements of an individual. Consider the images shown in Fig. 1. It is not very difficult to perceive the fact that these represent the silhouette of a walking human.

Apart from providing information about the activity being performed, the manner of shape changes provides valuable insights regarding the identity of the object. Gait-based human ID is an area that has attracted significant attention due to its potential applications in remote biometrics. The discrimination between individuals is significantly improved if we take the manner of shape changes into account.

In this chapter, we discuss approaches that specifically model both the appearance and behavior of a subject either using faces or using gait. In the case of face, the model parameters capture coarse facial appearance and global dynamic information. In the case of gait, the model captures the shape of the human silhouette and the observed variability in walking styles by non-linear warps of the time-axis.

1 Introduction

Biometrics refers to the measurement and analysis of unique physical or behavioral characteristics, especially as a means of verifying personal identity. Biometric systems work by using sensors to detect and perceive a certain physical characteristic. The system then converts them into digital patterns and compares these patterns with patterns stored for personal identification.

A lip is a tactile sensory organ constituting the visible portion of the mouth. The external-most skin of the lip is the stratified squamous epithelium. The surface of the lips has visible wrinkles that can be considered identifiable. This forms the foundation of Cheiloscopy, a field of forensics that focuses on identification of people by the use of lip traces.

Each lip print has a different pattern or texture of grooves on its surface. A lip print may not consist of only one type of groove alone but may have a mixture of varying types of patterns. The uniqueness of a lip has been proven by the researchers by using color information and shape analysis. These measures of lips along with the movement patterns during speech are unique for every person.

Lip patterns may be altered by external factors, but lips reassume their original pattern on recovery. Lip patterns are unique for individuals and remain unchanged during their lifetime, making them a reliable basis for identification.

This chapter proposes the use of deep learning, specifically convolutional neural networks (CNNs), to perform feature extraction and identification of lip images. The proposed approached uses CNNs to extract high-level spatiotemporal features from images of lips. This is achieved by training the neural network on images created from the frames of videos of people speaking. The minor variation of the shape and size of lips in the frames allow the neural network to identify changes with respect to speech.

The approach has been designed to work for grayscale images of lips, and the model has been able to achieve a significant level of accuracy. The proposed CNN model has been tested out on the OuluVS database [1], and the results obtained are presented in this chapter.

The addition of a cognitive aspect adds more information to the features obtained from behavioral biometric measures; this allows for an increase in the accuracy and effectiveness of the approach, specifically in the case of authentication systems.

Biometric imaging systems

Biometric imaging applications, such as fingerprint or iris recognition, provide the means to eliminate passwords, secure E-commerce, and securely identify humans or animals. In traditional imaging technology, quite often the trade-off between light throughput and depth of field makes conventional biometric systems difficult to use or not sufficiently reliable. By using Wavefront Coding, the depth of information gathered can greatly increasing the ease of use and possibly reducing the cost of the systems.

Figures 12a,b, and c shows example images of an iris taken with near IR illumination and a traditional CMOS imaging system at three different planes of focus. At best focus (Figure 12b) the detail of the iris is sharp and clear. This iris detail is used by iris recognition algorithms to accept or reject the person. With slight movement towards (Figure 12a) or away (Figure 12c) from the imaging system the detail of the iris is lost. Figures 12d,e, and f show example iris images taken with the same near-IR illumination and a CMOS Wavefront Coded imaging system. At best focus (Figure 12e), the iris detail is as high as that from the traditional system. Movement towards (Figure 12d) or away (Figure 12f) from the imaging system results in images with essentially a constant level of iris detail. The size of the iris changes through magnification change. Increasing the range at which clear images can be made extends the useful image capture volume, enabling a more flexible system that is easier to use.

14.1 Photoplethysmography in biometrics

14.1.2 Methods of PPG biometrics

Using physiological signals as a biometric is not as simple as the process of using fingerprints, where usually one image or high-resolution scan of the subject is required. Instead, the signal of interest, in this case the PPG, must go through a multistage process of acquisition, preprocessing and feature extraction. A database is then built that will be used as a template for matching in authentication scenarios. This method is well laid out in the review by Sancho et al. (2018). In their article the authors define that PPG biometric authentication involves two stages: (1) Enrollment, and (2) Matching.

Both stages have two common processing steps: PPG acquisition and PPG preprocessing. The later of these is split into four separate subprocesses, filtering, cycle detection, cycle normalization, and cycle alignment. The enrollment stage then creates a database, and the testing stage (used for authentication) will go through a matching step that compares the metrics of the acquired PPG signal with those in the pre-prepared database and will then make an access decision. The access decision itself will be the result of a predefined confidence interval where the PPG signal must achieve at least a minimum score on a set of matching criteria. The whole two-stage process is illustrated in Fig. 14.1.

Figure 14.1. Biometric process of enrollment and matching. After PPG acquisition, the signals are high-pass filtered (HPF) to remove the DC baseline caused by the strong absorption of the illuminating light source in the nonpulsatile tissue. The signal is then notch filtered (NF) to eliminate any power-line interference before a feature extraction and normalization routine is applied to the signal. The metrics calculated are either stored in or matched against the template database, and a decision is made to grant access based on a set of minimum matching criteria.

PPG acquisition, filtering, and feature extraction have been discussed in more detail in previous chapters. PPG feature extraction has been the subject of or part of the main work in many research articles (Elgendi, 2012; Reşit Kavsaoğlu et al., 2014). As discussed in these articles there are several main features of interest that can provide a unique dataset that can be used for example for blood pressure analysis (El Hajj and Kyriacou, 2020) and pulse rate variability (PRV) measurements (Mejía-Mejía et al., 2020).

The framework for PPG feature extraction involves first isolating each individual PPG pulse. The PPG waveform is defined in pulse intervals and is the distance between the lowest amplitude points before the onset of each systolic rise. Once the pulses have been isolated, features can be extracted and used for various applications as described above and in previous chapters. For biometric applications some of the main features that are utilized are listed below and illustrated in Fig. 14.2:

Figure 14.2. Main PPG signal features used for biometric applications. PPG signals are normalized in the time and amplitude dimensions before the systolic peak (S), diastolic Peak (DP), diachrotic notch (DN), and pulse width (PW) are identified. The timings of the various peaks and inflections are normally timed in normalized samples from the onset of the PPG waveform, called the foot (F). This figure shows two separate PPG pulses for illustrative purposes. In PPG feature extraction algorithms, all features are extracted from every pulse.

•

Systolic peak amplitude value and timing

•

Dicrotic notch amplitude value and timing

•

The diastolic peak amplitude and timing

•

Pulse width(s).

14.1.2.1 Systolic peak

This is the maximum amplitude of the pulse interval and provides a reference point for the normalization and alignment process. It also marks the portion of the PPG signal where the most amount of volume of blood has been detected at the PPG sensor (Allen, 2007).

14.1.2.2 Dicrotic notch

The point of inflection on the downslope of the PPG marking the end of the systolic phase before the onset of the diastolic peak.

14.1.2.3 Diastolic peak

The maximum amplitude in the diastolic portion of the PPG signal. The origin of the dicrotic notch and diastolic peak has been said to arise from pulse wave reflections at the aortoilliac junction (Li, 1985).

14.1.2.4 Pulse width

Various sources define different pulse widths and are normally denominated as the width at some percentage of the amplitude of the systolic peak. In their review of the relationship of systemic vascular resistance with the PPG, Awad et al. (2007) defined the pulse width as the width of the pulse at half the systolic peak amplitude.

Other features of the PPG signal that may be calculated and classified for the template database may include the timing between the systolic and diastolic peak (ΔT), pulse area (area under the curve between the pulse interval points), and inflection area ratio (the ratio between the area under the diastolic portion divided by the area under the systolic portion, marked by the dicrotic notch) first described by Wang et al. (2009).