**Physiological Signal Based Biometrics for Securing Body Sensor Network**

Fen Miao, Shu-Di Bao and Ye Li

Additional information is available at the end of the chapter

http://dx.doi.org/10.5772/51856

## **1. Introduction**

30 New Trends and Developments in Biometrics

250 New Trends and Developments in Biometrics

*Authentication (ICBA)*, pages 1-7.

[50] M. Indovina (2012), V. Dvornychenko, R. A. Hicklin, G. I. Kiebuzinski, ELFT-EFS Evaluation of Latent Fingerprint Technologies: Extended Feature Sets [Evaluation #2],

[51] D. Maio (2004), D. Maltoni, R. Cappelli, J. L. Wayman, A. K. Jain, FVC2004: Third Fingerprint Verification Competition, *Proc. International Conference on Biometric*

*NISTIR 7859*, http://dx.doi.org/10.6028/NIST.IR.7859

Nowadays, the constraints in the healthcare of developing countries, including high pop‐ ulation growth, a high burden of disease prevalence, low health care workforce, large numbers of rural inhabitants, and limited financial resources to support healthcare infra‐ structure and health information systems, accompanied with the improvement of poten‐ tial of lowering information and transaction costs in healthcare delivery due to the explosively access of mobile phones to all segments of a country, has motivated the de‐ velopment of mobile health or m-health field. M-health is known as the practice of medi‐ cal and public health supported by mobile devices such as mobile phones and PDAs for delivering medical and healthcare services. Thus, the popularity of m-health can be sub‐ jected to the development of wearable medical devices and wireless communication tech‐ nology. In order to fully utilize wireless technology between the wearable medical devices, the concept of body sensor network (BSN), which is a kind of wireless sensor network around human body, was proposed in 2002.

#### **1.1. Body sensor network**

BSN, which has great potential in being the main front-end platform of telemedicine and mobile health systems, is currently being heavily developed to keep pace with the continu‐ ously rising demand for personalized healthcare. Comprised of sensors attached to the hu‐ man body for collecting and transmitting vital signs, BSN is able to facilitate the joint processing of spatially and temporally collected medical data from different parts of the body for resource optimization and systematic health monitoring. In a typical BSN, each sensor node collects various physiological signals in order to monitor the patient's health status no matter their location and then instantly transmit all information in real time to the

© 2012 Miao et al.; licensee InTech. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. © 2012 Miao et al.; licensee InTech. This is a paper distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

medical server or the doctors. When an emergency is detected, the physicians will immedi‐ ately inform the patient through the computer system by providing appropriate messages or alarms. By this way, BSN is preferred in monitoring patients in environments lack of medi‐ cal doctors, such as home and workplaces. Fig.1 presents a simplified example of a BSN ap‐ plication scenario in a mobile health system. Sensor nodes on or inside the human body and a Master Node (MN), are connected to form a BSN. Medical information collected by differ‐ ent sensors in a BSN will be sent to the MN for data fusion and then to personal server for pre-processing before being forwarded to a central server for further analysis or the physi‐ cians for care giving via various forms of communications such as wireless personal area network (WPAN), wireless local area network (WLAN) and wide area network (WAN).

Symmetric cryptography, in which communication parties must possess a shared secret key via an invulnerable key distribution solution prior to any encryption process, is a promising approach to relieve the stringent resource constraints in BSN. Existing key distribution tech‐ niques for large-scale sensor networks, such as random-key pre-distribution protocols (Gli‐ gor et al, 2002; Perrig et al, 2003) and polynomial pool-based key distribution (Ning et al, 2003), require some form of pre-deployment. However, given the progressively increasing deployments of BSN, these approaches may potentially involve considerable latency during network initialization or any subsequent adjustments, due to their need for pre-deployment. In addition, it obviously discourages people, such as family members, to share sensors be‐ tween themselves because whenever there is need to add or change a body sensor, the user has to configure a new initial key to ensure that the new sensor can securely communicate with the existing ones. Therefore, a new series of key distribution solutions without any

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

253

form of initial deployment to provide plug and play security is desirable for BSNs.

As well known, the human body physiologically and biologically consists of its own transmission systems such as the blood circulation system, thus, how to make use of these secured communication pathways available specifically in BSN to secure it is a good idea (Poon et al, 2006). It is undoubtedly practical in securing BSN with a teleme‐ dicine or m-health application, as nodes of these BSN would already comprise biosen‐ sors for collecting medical data, which could be physiological characteristics uniquely representing an individual. If these intrinsic characteristics can be used to verify wheth‐ er two sensors belong to the same individual, the use of physiological signals to identify individuals and secure encryption key transmission with resources-saving is feasible. Building upon this initial idea, a family of lightweight and resource-efficient biometricsbased security solutions, which are based on time-variant physiological signals, has been proposed for the emerging BSN with a dual purpose of individual identification and key transmission. It is different from traditional biometrics, where the physiological or behavioural characteristics are static and merely used to automatic identify or verify an individual. The utilized biometric traits in traditional biometric systems should have the characteristics of universality, distinctiveness, permanence, effectiveness, invulnerability and so on, while the physiological characteristics should be dynamic at different times

As depicted in Fig.2, in biometrics solution the physiological signals of human body, such as electrocardiograph (ECG) and photoplethysmograph (PPG), were used to gener‐ ate the entity identifier (EI) of each node for identifying nodes and then protecting the transmission of keying materials by a key hiding/un-hiding process. It is based on the fact that EIs generated simultaneously from the same subject are with high similarity, while those generated non-simultaneously or from different subjects are with significant

**1.3. Novel biometrics for BSN security**

to ensure the security of key transmission in BSN.

differentiation.

**Figure 1.** An application scenario of BSN

#### **1.2. Security challenge in BSN**

As mandated by privacy laws and regulations, such as the Health Information and Portabili‐ ty Accountability Act (HIPAA) (Bowen et al, 2005) and the European Union Directive 2002/58/EC (2002), wireless standards with medical applications have to have a high level of reliability to guarantee the security of patients' information and the privacy of healthcare history. To ensure the security of the overall mobile health system, BSN as an important end, should be protected from different attacks such as eavesdropping, injection and modifi‐ cation. However, it is a nontrivial task due to stringently limited processing capability, memory, and energy, as well as lack of user interface, unskilled users, longevity of devices, and global roaming for most sensor nodes.

Symmetric cryptography, in which communication parties must possess a shared secret key via an invulnerable key distribution solution prior to any encryption process, is a promising approach to relieve the stringent resource constraints in BSN. Existing key distribution tech‐ niques for large-scale sensor networks, such as random-key pre-distribution protocols (Gli‐ gor et al, 2002; Perrig et al, 2003) and polynomial pool-based key distribution (Ning et al, 2003), require some form of pre-deployment. However, given the progressively increasing deployments of BSN, these approaches may potentially involve considerable latency during network initialization or any subsequent adjustments, due to their need for pre-deployment. In addition, it obviously discourages people, such as family members, to share sensors be‐ tween themselves because whenever there is need to add or change a body sensor, the user has to configure a new initial key to ensure that the new sensor can securely communicate with the existing ones. Therefore, a new series of key distribution solutions without any form of initial deployment to provide plug and play security is desirable for BSNs.

### **1.3. Novel biometrics for BSN security**

medical server or the doctors. When an emergency is detected, the physicians will immedi‐ ately inform the patient through the computer system by providing appropriate messages or alarms. By this way, BSN is preferred in monitoring patients in environments lack of medi‐ cal doctors, such as home and workplaces. Fig.1 presents a simplified example of a BSN ap‐ plication scenario in a mobile health system. Sensor nodes on or inside the human body and a Master Node (MN), are connected to form a BSN. Medical information collected by differ‐ ent sensors in a BSN will be sent to the MN for data fusion and then to personal server for pre-processing before being forwarded to a central server for further analysis or the physi‐ cians for care giving via various forms of communications such as wireless personal area network (WPAN), wireless local area network (WLAN) and wide area network (WAN).

Personal server

GPRS or 3G

As mandated by privacy laws and regulations, such as the Health Information and Portabili‐ ty Accountability Act (HIPAA) (Bowen et al, 2005) and the European Union Directive 2002/58/EC (2002), wireless standards with medical applications have to have a high level of reliability to guarantee the security of patients' information and the privacy of healthcare history. To ensure the security of the overall mobile health system, BSN as an important end, should be protected from different attacks such as eavesdropping, injection and modifi‐ cation. However, it is a nontrivial task due to stringently limited processing capability, memory, and energy, as well as lack of user interface, unskilled users, longevity of devices,

WAN

BSN BSN BSN

Respiration ECGI

ECGII

**Figure 1.** An application scenario of BSN

**1.2. Security challenge in BSN**

and global roaming for most sensor nodes.

EEG

252 New Trends and Developments in Biometrics

MN

PPG

Blood Pressure

> Medical Server

Professional

Base station

As well known, the human body physiologically and biologically consists of its own transmission systems such as the blood circulation system, thus, how to make use of these secured communication pathways available specifically in BSN to secure it is a good idea (Poon et al, 2006). It is undoubtedly practical in securing BSN with a teleme‐ dicine or m-health application, as nodes of these BSN would already comprise biosen‐ sors for collecting medical data, which could be physiological characteristics uniquely representing an individual. If these intrinsic characteristics can be used to verify wheth‐ er two sensors belong to the same individual, the use of physiological signals to identify individuals and secure encryption key transmission with resources-saving is feasible. Building upon this initial idea, a family of lightweight and resource-efficient biometricsbased security solutions, which are based on time-variant physiological signals, has been proposed for the emerging BSN with a dual purpose of individual identification and key transmission. It is different from traditional biometrics, where the physiological or behavioural characteristics are static and merely used to automatic identify or verify an individual. The utilized biometric traits in traditional biometric systems should have the characteristics of universality, distinctiveness, permanence, effectiveness, invulnerability and so on, while the physiological characteristics should be dynamic at different times to ensure the security of key transmission in BSN.

As depicted in Fig.2, in biometrics solution the physiological signals of human body, such as electrocardiograph (ECG) and photoplethysmograph (PPG), were used to gener‐ ate the entity identifier (EI) of each node for identifying nodes and then protecting the transmission of keying materials by a key hiding/un-hiding process. It is based on the fact that EIs generated simultaneously from the same subject are with high similarity, while those generated non-simultaneously or from different subjects are with significant differentiation.

detection in FDPS-based EI generation scheme. However, the poor randomness and recogni‐ tion rate performance are the bottlenecks of using FDPS to generate EIs and need to be bro‐

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

255

Since intrinsic characteristics captured simultaneously at different parts of the same subject have slight differences, fuzzy methods should be deployed on the transmitter/receiver for an increased tolerance in acceptable differences to protect the transmission of keying materi‐ als using generated EIs. Fuzzy commitment scheme proposed by Juels (2002), which works effectively in the case that the generated EIs are all sequential and with the same length, is employed in BSN security due to its low computational complexity, low memory occupied, as well as convenience to be implemented. However, Fuzzy commitment scheme is not ap‐ propriate while the feature points in EIs are un-ordered or with missing values due to its requirement for correspondence of features in terms of order. To address this issue, Juels and Sudan (2006) proposed the fuzzy vault scheme, which offers attractive properties in terms of security, changeable key, and flexibility, and thus has been a good candidate for biometrics based cryptographic systems. It has been applied in different traditional biomet‐ ric systems, for example, fingerprint, face, and Iris biometric systems for better performance than fuzzy commitment. Though fuzzy vault scheme was also adopted in biometrics based BSN security in more and more studies, it is noted that (Miao et al, 2010) the scheme is not good enough to achieve stable performance if the generated EIs are with dynamic random patterns in bit difference. Also, fuzzy vault has its drawbacks of low recognition rate due to not considering the inequality of the number of features in EIs generated from the two com‐

This chapter will describe the aspects of this kind of new biometrics with focus on the stateof-the-art biometric solutions for BSN security. In Section 2, the schemes of generating EIs from physiological signals based on both time-domain and frequency-domain information will be presented, followed by the performance evaluation as being a dynamic individual identifier to differentiate different subjects. Secondly, the usage of such generated EIs for se‐ curing BSN, i.e. key transmission schemes, will be detailed with a performance comparison of different schemes designed according to EIs' specific characteristics in Section 3. In Sec‐ tion 4, we conclude this chapter with an illustration of different biometric solutions in BSN

EI generation scheme is the most important issue should be addressed in the biometrics sol‐ utions because the security of BSN depends heavily on the characteristics of EIs generated. As described in Section 1, the state-of-the-art EI generation schemes are mainly classified in‐ to two categories, one is based on the time-domain information of physiological signals (TDPS) and the other is based on the frequency-domain information (FDPS). In this section,

security, where some issues need to be further studied will be emphasized.

**2. Entity identifier generation schemes**

ken through to ensure the security performance of BSN.

*1.3.2. Fuzzy method based key distribution solution*

munication parties.

**Figure 2.** Workflow of biometrics-based security solution

#### *1.3.1. Dynamic EI generation*

The timing information of heartbeats was demonstrated by Bao et al (2005) to be a possible biometric characteristic to be used in proposed entity authentication scheme due to its cha‐ otic nature, which can ensure the dynamic random performance of the generated EIs and then the security performance for BSN. Thus, the authors proposed to use Inter-Pulse-Inter‐ val (IPI) to generate EIs for securing the distribution of keying materials. A rigorously infor‐ mation-theoretic secure extraction scheme to properly extract the randomness of ECG signal, mainly from the IPI information, was proposed by Xu et al (2011). It was demonstrat‐ ed that there are two advantages of using IPI to secure BSN. Firstly, it can be derived from multiple physiological signals such as electrocardiograph (ECG) and photoplethysmograph (PPG) by measuring the time difference between peaks in the signals. Secondly, it has been demonstrated that EIs generated from a series of IPI values passed the selected randomness tests from the National Institute of Standards and Technology (NIST) standards, and thus show an acceptable degree of randomness. However, an R-wave detection process is re‐ quired before IPI measurement, which not only increases the computational complexity, but also leads to the uncertain performance because the accuracy of R-wave detection seriously affects the performance of IPI-based security system. In addition, 32 IPIs need to be utilized to generate a 128-bit EI, which means about 30 seconds of ECG/PPG measurements are re‐ quired before cryptographic keys can be securely distributed. To overcome these problems, the frequency-domain characteristics of physiological signals (FDPS), was proposed by Gup‐ ta et al (2010) to be a promising biometric characteristic due to its real-time performance, where 5 seconds measurement is enough to generate EIs. Also, there is no need of R-wave detection in FDPS-based EI generation scheme. However, the poor randomness and recogni‐ tion rate performance are the bottlenecks of using FDPS to generate EIs and need to be bro‐ ken through to ensure the security performance of BSN.

## *1.3.2. Fuzzy method based key distribution solution*

Dynamic EI generation

Physiological Signal

*K*

254 New Trends and Developments in Biometrics

**Figure 2.** Workflow of biometrics-based security solution

*1.3.1. Dynamic EI generation*

Transmitter

Receiver

Dynamic EI generation

Key Un-hiding

Physiological Signal

*K*

Y

fail

N

Verification

*EI EI'*

The timing information of heartbeats was demonstrated by Bao et al (2005) to be a possible biometric characteristic to be used in proposed entity authentication scheme due to its cha‐ otic nature, which can ensure the dynamic random performance of the generated EIs and then the security performance for BSN. Thus, the authors proposed to use Inter-Pulse-Inter‐ val (IPI) to generate EIs for securing the distribution of keying materials. A rigorously infor‐ mation-theoretic secure extraction scheme to properly extract the randomness of ECG signal, mainly from the IPI information, was proposed by Xu et al (2011). It was demonstrat‐ ed that there are two advantages of using IPI to secure BSN. Firstly, it can be derived from multiple physiological signals such as electrocardiograph (ECG) and photoplethysmograph (PPG) by measuring the time difference between peaks in the signals. Secondly, it has been demonstrated that EIs generated from a series of IPI values passed the selected randomness tests from the National Institute of Standards and Technology (NIST) standards, and thus show an acceptable degree of randomness. However, an R-wave detection process is re‐ quired before IPI measurement, which not only increases the computational complexity, but also leads to the uncertain performance because the accuracy of R-wave detection seriously affects the performance of IPI-based security system. In addition, 32 IPIs need to be utilized to generate a 128-bit EI, which means about 30 seconds of ECG/PPG measurements are re‐ quired before cryptographic keys can be securely distributed. To overcome these problems, the frequency-domain characteristics of physiological signals (FDPS), was proposed by Gup‐ ta et al (2010) to be a promising biometric characteristic due to its real-time performance, where 5 seconds measurement is enough to generate EIs. Also, there is no need of R-wave

Synchronization

Key Hiding

Since intrinsic characteristics captured simultaneously at different parts of the same subject have slight differences, fuzzy methods should be deployed on the transmitter/receiver for an increased tolerance in acceptable differences to protect the transmission of keying materi‐ als using generated EIs. Fuzzy commitment scheme proposed by Juels (2002), which works effectively in the case that the generated EIs are all sequential and with the same length, is employed in BSN security due to its low computational complexity, low memory occupied, as well as convenience to be implemented. However, Fuzzy commitment scheme is not ap‐ propriate while the feature points in EIs are un-ordered or with missing values due to its requirement for correspondence of features in terms of order. To address this issue, Juels and Sudan (2006) proposed the fuzzy vault scheme, which offers attractive properties in terms of security, changeable key, and flexibility, and thus has been a good candidate for biometrics based cryptographic systems. It has been applied in different traditional biomet‐ ric systems, for example, fingerprint, face, and Iris biometric systems for better performance than fuzzy commitment. Though fuzzy vault scheme was also adopted in biometrics based BSN security in more and more studies, it is noted that (Miao et al, 2010) the scheme is not good enough to achieve stable performance if the generated EIs are with dynamic random patterns in bit difference. Also, fuzzy vault has its drawbacks of low recognition rate due to not considering the inequality of the number of features in EIs generated from the two com‐ munication parties.

This chapter will describe the aspects of this kind of new biometrics with focus on the stateof-the-art biometric solutions for BSN security. In Section 2, the schemes of generating EIs from physiological signals based on both time-domain and frequency-domain information will be presented, followed by the performance evaluation as being a dynamic individual identifier to differentiate different subjects. Secondly, the usage of such generated EIs for se‐ curing BSN, i.e. key transmission schemes, will be detailed with a performance comparison of different schemes designed according to EIs' specific characteristics in Section 3. In Sec‐ tion 4, we conclude this chapter with an illustration of different biometric solutions in BSN security, where some issues need to be further studied will be emphasized.

## **2. Entity identifier generation schemes**

EI generation scheme is the most important issue should be addressed in the biometrics sol‐ utions because the security of BSN depends heavily on the characteristics of EIs generated. As described in Section 1, the state-of-the-art EI generation schemes are mainly classified in‐ to two categories, one is based on the time-domain information of physiological signals (TDPS) and the other is based on the frequency-domain information (FDPS). In this section, we will illustrate the two schemes separately with a detail performance evaluation on their advantages and disadvantages.

#### **2.1. TDPS-based EI generation scheme**

IPI is the most commonly used timing information in TDPS-based EI generation scheme. Fig.3 presents the experimental protocol of IPI-based EI generation scheme and the applica‐ tion of EIs for node identification. In IPI-based EI generation scheme, each node extracts the time-domain information by calculating a series of IPIs from its own recorded cardiovascu‐ lar signal such as ECG and PPG based on a synchronization signal initiated by the master node, which can be denoted as {*IPIi* |1≤*i* ≤ *N* }. IPI-based EI generation process is then de‐ ployed on the series of IPIs of each end to generate its own EI. The EIs generated simultane‐ ously from the transmitter and the receiver are with high similarity for the same subject, while high dissimilarity for different subjects or generated non-simultaneously, and thus can be used to identify nodes by comparing the Hamming distance between two EIs.

As depicted in Fig.3, given a sequence of IPI values, the IPI-based EI generation process for the transmitter/receiver is comprised of the following three processes: accumulation & mod‐ ulo, contraction mapping and Gray coding. Give *N* consecutive individual IPIs, a series of multi-IPIs can be obtained as follows:

$$\left\{ mIPI\_i = \sum\_{n=1}^{i} IPI\_n \; \middle| \; 1 \le i \le N \right\} \tag{1}$$

*f* ^

where *L* >*q* and . returns the largest integer less than or equal to *<sup>m</sup>*

from a corresponding *mIPIi*

**2.2. FDPS-based EI generation scheme**

which will be explained in Section 2.3.

*Signal Acquisition*

*N* ×*q*.

EIs. The generated EI can be expressed as *EI* = *I*<sup>1</sup> | | *I*2⋯ | | *I <sup>L</sup>* <sup>−</sup><sup>1</sup> | | *IN* , where *Ii*

(*m*)= *<sup>m</sup>*

crease the noise margin of measurements, the Gray code scheme is employed to get binary

Fig.4 presents a demonstration of the experimental protocol of FDPS-based EI generation scheme and the application of EIs for node identification with PPG as the physiological sig‐ nal. In state-of-the-art FDPS-based EI generation schemes, nodes in the same BSN obtained independently the same physiological signal in a loosely synchronized manner, at a specific sampling rate for a fixed duration. An EI generation process is then deployed on the signal acquired from each end to generate its own EI. In order to realize node identification and the security of keying materials, the EIs generated simultaneously from the transmitter and the receiver should be with high similarity for the same subject, while high dissimilarity for dif‐ ferent subjects or generated non-simultaneously. Therefore, the EIs can be used to identify nodes by comparing the distance between two EIs. Different from TDPS-based EI generation scheme, the distance of EIs measured here cannot be Hamming distance, the reason of

<sup>2</sup>(*<sup>L</sup>* <sup>−</sup>*q*) (2)

Physiological Signal Based Biometrics for Securing Body Sensor Network

with the bit length of *q*. Such generated EIs have a bit length of

2(*<sup>L</sup>* <sup>−</sup>*q*)

http://dx.doi.org/10.5772/51856

matched Y

Distance< threshold

*Process*

N

Unmatched

DTW Distance

*Process Node Identification* 

EI

FFT

*EI Generation* 

Peak Detection

Randomized Algorithm

> Binary Encoding

As depicted in Fig.4, an entire FDPS-based EI generation process is comprised of a Fast Fourier Transform (FFT) process, a peak detection process, a randomized algorithm in some situations and a binary encoding process. In the previous FDPS-based EI generation process proposed by Gupta *et al* (2010), the samples collected are divided into several windows and

**Figure 4.** Experimental Protocol of FDPS-based EI generation scheme and node identification

EI'

. Finally, to in‐

257

is generated

**Figure 3.** Experimental Protocol of TDPS-based EI generation scheme and node identification

To randomize the monotonically increasing multi-IPIs, a modulo operation is further intro‐ duced, i.e. (*mIPIi* )mod(2*<sup>L</sup>* ), where *L* is a positive integer referred to as the modulo parame‐ ter. To compensate measurement differences among different BSN nodes, the modulo result is further transformed into a small integer *q* by a contraction mapping *f* ^ : 0, <sup>2</sup>*<sup>L</sup>* )<sup>→</sup> 0, <sup>2</sup>*<sup>q</sup>* ), i.e.,

$$\stackrel{\frown}{f}(m) = \left\lfloor \frac{m}{2^{(1-\gamma)}} \right\rfloor \tag{2}$$

where *L* >*q* and . returns the largest integer less than or equal to *<sup>m</sup>* 2(*<sup>L</sup>* <sup>−</sup>*q*) . Finally, to in‐ crease the noise margin of measurements, the Gray code scheme is employed to get binary EIs. The generated EI can be expressed as *EI* = *I*<sup>1</sup> | | *I*2⋯ | | *I <sup>L</sup>* <sup>−</sup><sup>1</sup> | | *IN* , where *Ii* is generated from a corresponding *mIPIi* with the bit length of *q*. Such generated EIs have a bit length of *N* ×*q*.

#### **2.2. FDPS-based EI generation scheme**

we will illustrate the two schemes separately with a detail performance evaluation on their

IPI is the most commonly used timing information in TDPS-based EI generation scheme. Fig.3 presents the experimental protocol of IPI-based EI generation scheme and the applica‐ tion of EIs for node identification. In IPI-based EI generation scheme, each node extracts the time-domain information by calculating a series of IPIs from its own recorded cardiovascu‐ lar signal such as ECG and PPG based on a synchronization signal initiated by the master node, which can be denoted as {*IPIi* |1≤*i* ≤ *N* }. IPI-based EI generation process is then de‐ ployed on the series of IPIs of each end to generate its own EI. The EIs generated simultane‐ ously from the transmitter and the receiver are with high similarity for the same subject, while high dissimilarity for different subjects or generated non-simultaneously, and thus

can be used to identify nodes by comparing the Hamming distance between two EIs.

{IPI1, IPI2, …, IPIN}

{IPI'1, IPI'2, …, IPI'N}

To randomize the monotonically increasing multi-IPIs, a modulo operation is further intro‐

ter. To compensate measurement differences among different BSN nodes, the modulo result is

{*mIPIi* <sup>=</sup>∑ *n*=1 *i*

*IPI Extraction Process EI Generation* 

**Figure 3.** Experimental Protocol of TDPS-based EI generation scheme and node identification

further transformed into a small integer *q* by a contraction mapping *f*

As depicted in Fig.3, given a sequence of IPI values, the IPI-based EI generation process for the transmitter/receiver is comprised of the following three processes: accumulation & mod‐ ulo, contraction mapping and Gray coding. Give *N* consecutive individual IPIs, a series of

*IPIn* |1≤*i* ≤ *N* } (1)

Hamming Distance

EI

Accumulation & Modulo

*Process*

Contraction mappring

Gray coding

)mod(2*<sup>L</sup>* ), where *L* is a positive integer referred to as the modulo parame‐

EI'

matched Y

^ : 0, <sup>2</sup>*<sup>L</sup>* )<sup>→</sup> 0, <sup>2</sup>*<sup>q</sup>*

Distance< threshold

*Node Identification Process*

> N Unmatched

> > ), i.e.,

advantages and disadvantages.

256 New Trends and Developments in Biometrics

**2.1. TDPS-based EI generation scheme**

multi-IPIs can be obtained as follows:

IPI1 IPI2 IPIN

IPI'1 IPI'2 IPI'N

duced, i.e. (*mIPIi*

Fig.4 presents a demonstration of the experimental protocol of FDPS-based EI generation scheme and the application of EIs for node identification with PPG as the physiological sig‐ nal. In state-of-the-art FDPS-based EI generation schemes, nodes in the same BSN obtained independently the same physiological signal in a loosely synchronized manner, at a specific sampling rate for a fixed duration. An EI generation process is then deployed on the signal acquired from each end to generate its own EI. In order to realize node identification and the security of keying materials, the EIs generated simultaneously from the transmitter and the receiver should be with high similarity for the same subject, while high dissimilarity for dif‐ ferent subjects or generated non-simultaneously. Therefore, the EIs can be used to identify nodes by comparing the distance between two EIs. Different from TDPS-based EI generation scheme, the distance of EIs measured here cannot be Hamming distance, the reason of which will be explained in Section 2.3.

**Figure 4.** Experimental Protocol of FDPS-based EI generation scheme and node identification

As depicted in Fig.4, an entire FDPS-based EI generation process is comprised of a Fast Fourier Transform (FFT) process, a peak detection process, a randomized algorithm in some situations and a binary encoding process. In the previous FDPS-based EI generation process proposed by Gupta *et al* (2010), the samples collected are divided into several windows and a FFT is performed on each of these parts, denoted as Multi-Windows generation scheme. A combination with the form of < *Kx i* , *Ky i* > is derived through the peak detection process de‐ ployed on the FFT coefficients, where *Kx i* is the FFT point at which peak is observed, *Ky i* is the corresponding FFT coefficient values, and *i* is the index of the peaks. Each of the peakindex and peak-value are quantized and converted into a binary string and concatenated to form an EI, which can be denoted as *EI* ={ *f* <sup>1</sup> , *f* <sup>2</sup> , ⋯ *f <sup>N</sup>* }, where *f <sup>i</sup>* = *Kx i* , *Ky <sup>i</sup>* , 1≤*i* ≤ *N* , *N* is the number of indexes where peaks are observed, which varies upon situation. However, based on what we learned from experimental analysis, *Ky <sup>i</sup>* is not a good resource to generate EIs because the amplitudes of physiological signals can be easily affected by a lot of meas‐ urement factors, such as the degree of skin exposure to sensor nodes.

Therefore, a Single-Window (SW) method to generate EIs was proposed by Miao *et al* (2011) aiming for a significant improvement in recognition performance and increase in random‐ ness performance. Firstly, FFT is directly performed on the physiological signal in a loosely synchronized manner, at a specific sampling rate for a fixed duration, such as 5 seconds.

indexes where peaks are observed, which varies upon situation. Before binary encoding process, a randomized algorithm is designed to overcome the bottleneck of randomness per‐ formance. Fig.5(a) and Fig.5(b) indicate an example of the differences between Multi-Win‐ dows (MWs) and Single-Window (SW) methods before randominzed algorithm with ECG signals collected by two nodes on one single subject and from two different subjects, respec‐ tively, while Fig.5(c) and Fig.5(d) indicate an example of the differences with PPG signals. It can be seen from Fig.5 that the number of peaks generated from nodes on the same subject has a higher number of matchings in terms of peak-index compared to those from different subjects; however, there is no such findings with FFT coefficients. In addtion, compared with MWs method, the SW method presents a larger matching rate, which is defined as the rate between the number of matched peaks and the number of the detected peaks, for the same subject and smaller matching rate for different subjects, no matter what kind of phys‐

Obviously, all of the integer values of feature points are ascending and within a certain scale, which would bring about the bottleneck of the randomness performance and security weakness. Therefore, a randomized algorithm similar to Linear Congruential Generator (LCG), which is a kind of pseudorandom number generator, is deployed to randomize *F*

<sup>2</sup> + *c*)mod2*<sup>p</sup>*

the randomness performance of *F*. In the randomized algorithm, it is recommended that the

where *β* ≥0, x returns the largest integer less than or equal to x. Then, a permuted feature points set is generated with the form of

2, ⋯ *f* ′

*<sup>c</sup>* =2*<sup>β</sup>* <sup>+</sup> 1, *<sup>c</sup>* / <sup>2</sup>*<sup>p</sup>* =(1 / <sup>2</sup><sup>−</sup> <sup>3</sup>

1, *f* ′

, ⋯(*bKx*

6 )

is the "increment". The selection of *b*, *c*, *p* is directly related to

={ *<sup>f</sup>* 1, *<sup>f</sup>* <sup>2</sup>⋯, *<sup>f</sup> <sup>N</sup>* } (3)

is the "modulus" and *p* is a positive integer referred to as modulo parameter, *b* ≥0

*<sup>N</sup>* + *c*)mod2*p*}

*<sup>N</sup>* ) by randomly permuting the order of

(4)

1 , *Kx* 2 , ⋯ *Kx*

>, of the first *M* points of FFT coefficients is selected and

Physiological Signal Based Biometrics for Securing Body Sensor Network

*<sup>N</sup>* }, where *N* is the number of

http://dx.doi.org/10.5772/51856

259

Then, each peak-index, i.e. < *Kx*

iological signal is.

and form a new set *F* ′

where 2*<sup>p</sup>*

*F* ′

is the "multiplier", 0≤*c* <2*<sup>p</sup>*

={(*bKx*

*<sup>F</sup>* ″ <sup>=</sup>*RandomPermute*( *<sup>f</sup>* 1, *<sup>f</sup>* <sup>2</sup>, <sup>⋯</sup>, *<sup>f</sup> <sup>N</sup>* )=( *<sup>f</sup>* ′

, i.e.,

<sup>1</sup> + *c*)mod2*<sup>p</sup>*

most optimal relationship between *b*, *c*, *p* is as followings:

{

*b* = 2*p*/2 + 1

, (*bKx*

*i*

concatenated to form a set of feature points *F* ={*Kx*

**Figure 5.** Multi-Windows versus Single-Window feature points generation process: (a) Same subject with ECG; (b) Dif‐ ferent subjects with ECG; (c) Same subject with PPG; (d) Different subjects with PPG

Therefore, a Single-Window (SW) method to generate EIs was proposed by Miao *et al* (2011) aiming for a significant improvement in recognition performance and increase in random‐ ness performance. Firstly, FFT is directly performed on the physiological signal in a loosely synchronized manner, at a specific sampling rate for a fixed duration, such as 5 seconds. Then, each peak-index, i.e. < *Kx i* >, of the first *M* points of FFT coefficients is selected and concatenated to form a set of feature points *F* ={*Kx* 1 , *Kx* 2 , ⋯ *Kx <sup>N</sup>* }, where *N* is the number of indexes where peaks are observed, which varies upon situation. Before binary encoding process, a randomized algorithm is designed to overcome the bottleneck of randomness per‐ formance. Fig.5(a) and Fig.5(b) indicate an example of the differences between Multi-Win‐ dows (MWs) and Single-Window (SW) methods before randominzed algorithm with ECG signals collected by two nodes on one single subject and from two different subjects, respec‐ tively, while Fig.5(c) and Fig.5(d) indicate an example of the differences with PPG signals. It can be seen from Fig.5 that the number of peaks generated from nodes on the same subject has a higher number of matchings in terms of peak-index compared to those from different subjects; however, there is no such findings with FFT coefficients. In addtion, compared with MWs method, the SW method presents a larger matching rate, which is defined as the rate between the number of matched peaks and the number of the detected peaks, for the same subject and smaller matching rate for different subjects, no matter what kind of phys‐ iological signal is.

a FFT is performed on each of these parts, denoted as Multi-Windows generation scheme. A

the corresponding FFT coefficient values, and *i* is the index of the peaks. Each of the peakindex and peak-value are quantized and converted into a binary string and concatenated to

, *f* <sup>2</sup>

the number of indexes where peaks are observed, which varies upon situation. However,

EIs because the amplitudes of physiological signals can be easily affected by a lot of meas‐

**Figure 5.** Multi-Windows versus Single-Window feature points generation process: (a) Same subject with ECG; (b) Dif‐

ferent subjects with ECG; (c) Same subject with PPG; (d) Different subjects with PPG

*i*

> is derived through the peak detection process de‐

is the FFT point at which peak is observed, *Ky*

= *Kx i* , *Ky*

*<sup>i</sup>* is not a good resource to generate

, ⋯ *f <sup>N</sup>* }, where *f <sup>i</sup>*

*i* is

*<sup>i</sup>* , 1≤*i* ≤ *N* , *N* is

*i* , *Ky i*

combination with the form of < *Kx*

258 New Trends and Developments in Biometrics

ployed on the FFT coefficients, where *Kx*

form an EI, which can be denoted as *EI* ={ *f* <sup>1</sup>

based on what we learned from experimental analysis, *Ky*

urement factors, such as the degree of skin exposure to sensor nodes.

Obviously, all of the integer values of feature points are ascending and within a certain scale, which would bring about the bottleneck of the randomness performance and security weakness. Therefore, a randomized algorithm similar to Linear Congruential Generator (LCG), which is a kind of pseudorandom number generator, is deployed to randomize *F* and form a new set *F* ′ , i.e.,

$$\begin{aligned} \text{If } F &= \left\{ (bK\_x^1 + c) \text{mod}\, \mathbf{2}^p \, , (bK\_x^2 + c) \text{mod}\, \mathbf{2}^p \, \cdots \, (bK\_x^N + c) \text{mod}\, \mathbf{2}^p \right\} \\ &= \left\{ f\_1 \, \, f\_2 \, \cdots \, \, f\_N \right\} \end{aligned} \tag{3}$$

where 2*<sup>p</sup>* is the "modulus" and *p* is a positive integer referred to as modulo parameter, *b* ≥0 is the "multiplier", 0≤*c* <2*<sup>p</sup>* is the "increment". The selection of *b*, *c*, *p* is directly related to the randomness performance of *F*. In the randomized algorithm, it is recommended that the most optimal relationship between *b*, *c*, *p* is as followings:

$$\begin{cases} b = \lfloor 2^{p/2} \rfloor + 1 \\ c = 2\beta + 1, \ c \left\lfloor 2^p = (1 \mid 2 - \sqrt{5} \right\rfloor \right\rfloor \end{cases} \tag{4}$$

where *β* ≥0, x returns the largest integer less than or equal to x. Then, a permuted feature points set is generated with the form of *<sup>F</sup>* ″ <sup>=</sup>*RandomPermute*( *<sup>f</sup>* <sup>1</sup>, *<sup>f</sup>* <sup>2</sup>, <sup>⋯</sup>, *<sup>f</sup> <sup>N</sup>* )=( *<sup>f</sup>* ′ 1, *f* ′ 2, ⋯ *f* ′ *<sup>N</sup>* ) by randomly permuting the order of each point *f <sup>i</sup>* . The generated EI can be expressed as *EI* = *I*<sup>1</sup> | | *I*2⋯ | | *I <sup>N</sup>* <sup>−</sup><sup>1</sup> | | *IN* , where | | is a concatenation operation. Each block of EI, i.e., *Ii* is the binary result of a corresponding *f <sup>i</sup>* ′. The bit length of *Ii* is *p*, and thus, the bit length of EI is *N* × *p*.

FDPS-based EIs passed the selected tests, and TDPS-based EIs show a better randomness

The similarity between any pair of TDPS-based EIs generated simultaneously by sensors on the same individual can be analyzed with the Hamming distance. Fig.6 depicts the Ham‐ ming distance distribution of EIs with *L* =8, *N* =16, *q* =3. It can be seen that more than 95% of the Hamming distances between TDPS-based EIs are less than 10, and thus shows a good group similarity performance with the two experiments. Therefore, the proposed scheme is

> Exp. I Exp. II

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

261

0 10 20 30 40

Hamming distance

Different from TDPS-based EIs, FDPS-based EIs cannot be analyzed with the Hamming dis‐ tance due to the unequal length of generated EIs and matching points at different orders. As shown in Fig.7, the feature sets generated at the transmitter/receiver of the same subject after randomized algorithm have some common points, such as 174, 225, 156 at different orders, and thus direct Hamming distance in sequence can not reflect the real matching performance.

106 174 225 3 71 156 241 36 121 171 1

174 225 20 88 156 241 121 155 1 86 171 0

performance than FDPS-based EIs.

applicable in both healthy people and clinical subjects.

0

**Figure 6.** Similarity analysis with the Hamming distance (*L* =8, *N* =16, *q* =3)

**Figure 7.** Feature points set generated at the transmitter/receiver of the same subject

0.05

0.1

0.15

Probability

0.2

0.25

*2.3.2. Group similarity analysis*

#### **2.3. Performance evaluation**

To demonstrate the performance of different EI generation scheme, we conduct the perform‐ ance evaluation in terms of randomness performance and group similarity. The perform‐ ance comparison will be given to systematically illustrate the advantages and disadvantages of different schemes with two experiments. In the first experiment (Exp. I), the experimental data to be used for performance evaluation include ECG and PPG from 14 healthy subjects, where ECG was captured from the three fingers of each subjects and two channels of PPG were captured from the index fingers of the two hands, respectively. For each subject, the three channels of signals were captured simultaneously for 2-3min. All the three channels of signals were used to generate TDPS-based EIs, while two channels of PPG were used to gen‐ erate FDPS-based EIs. In the second experiment (Exp. II), there were in total 85 clinical sub‐ jects from the hospital and two channels of physiological signals (including one-channel ECG and one-channel PPG) with a duration of 40 seconds were simultaneously recorded from each subject on three or four days within two-month period.

#### *2.3.1. Randomness performance analysis*

The randomness performance of binary sequences can be evaluated using a variety of ran‐ domness tests. Beacause of the length limitation in the generated binary EIs from each sub‐ ject, several tests from the National Institute of Standards and Technology (NIST) standards were selected, including frequency (monobit) test, frequency test within a block, cumulative sums test, runs test, and approximate entropy test with the decision rule of 1% level.


**Table 1.** Randomness test results

Table 1 shows the randomness test results of TDPS-based EIs and FDPS-based EIs, where the randomizing parameters in FDPS-based EI generation scheme were set as *b* =23, *c* =109, *p* =9. It can be seen that all bit streams generated based on TDPS and most of FDPS-based EIs passed the selected tests, and TDPS-based EIs show a better randomness performance than FDPS-based EIs.

#### *2.3.2. Group similarity analysis*

each point *f <sup>i</sup>*

The bit length of *Ii*

**2.3. Performance evaluation**

260 New Trends and Developments in Biometrics

*2.3.1. Randomness performance analysis*

**Test**

**Table 1.** Randomness test results

. The generated EI can be expressed as *EI* = *I*<sup>1</sup> | | *I*2⋯ | | *I <sup>N</sup>* <sup>−</sup><sup>1</sup> | | *IN* , where | | is

′.

a concatenation operation. Each block of EI, i.e., *Ii* is the binary result of a corresponding *f <sup>i</sup>*

To demonstrate the performance of different EI generation scheme, we conduct the perform‐ ance evaluation in terms of randomness performance and group similarity. The perform‐ ance comparison will be given to systematically illustrate the advantages and disadvantages of different schemes with two experiments. In the first experiment (Exp. I), the experimental data to be used for performance evaluation include ECG and PPG from 14 healthy subjects, where ECG was captured from the three fingers of each subjects and two channels of PPG were captured from the index fingers of the two hands, respectively. For each subject, the three channels of signals were captured simultaneously for 2-3min. All the three channels of signals were used to generate TDPS-based EIs, while two channels of PPG were used to gen‐ erate FDPS-based EIs. In the second experiment (Exp. II), there were in total 85 clinical sub‐ jects from the hospital and two channels of physiological signals (including one-channel ECG and one-channel PPG) with a duration of 40 seconds were simultaneously recorded

The randomness performance of binary sequences can be evaluated using a variety of ran‐ domness tests. Beacause of the length limitation in the generated binary EIs from each sub‐ ject, several tests from the National Institute of Standards and Technology (NIST) standards were selected, including frequency (monobit) test, frequency test within a block, cumulative

**Pass rate**

**FDPS-based EIs** *b***=23,** *c* **=109,** *p* **=9 TDPS-based EIs**

sums test, runs test, and approximate entropy test with the decision rule of 1% level.

Frequency Test 99.219% 100% Frequency Test within a Block (M=10) 100% 100% Runs Test 99.219% 100% Cumulative Sums Test 100% 100% Approximate Entropy Test 99.219% 100%

Table 1 shows the randomness test results of TDPS-based EIs and FDPS-based EIs, where the randomizing parameters in FDPS-based EI generation scheme were set as *b* =23, *c* =109, *p* =9. It can be seen that all bit streams generated based on TDPS and most of

is *p*, and thus, the bit length of EI is *N* × *p*.

from each subject on three or four days within two-month period.

The similarity between any pair of TDPS-based EIs generated simultaneously by sensors on the same individual can be analyzed with the Hamming distance. Fig.6 depicts the Ham‐ ming distance distribution of EIs with *L* =8, *N* =16, *q* =3. It can be seen that more than 95% of the Hamming distances between TDPS-based EIs are less than 10, and thus shows a good group similarity performance with the two experiments. Therefore, the proposed scheme is applicable in both healthy people and clinical subjects.

**Figure 6.** Similarity analysis with the Hamming distance (*L* =8, *N* =16, *q* =3)

Different from TDPS-based EIs, FDPS-based EIs cannot be analyzed with the Hamming dis‐ tance due to the unequal length of generated EIs and matching points at different orders. As shown in Fig.7, the feature sets generated at the transmitter/receiver of the same subject after randomized algorithm have some common points, such as 174, 225, 156 at different orders, and thus direct Hamming distance in sequence can not reflect the real matching performance.

**Figure 7.** Feature points set generated at the transmitter/receiver of the same subject

Therefore, dynamic time warping (DTW) distance was selected to measure the group simi‐ larity between any pair of EIs generated from the same subject. DTW is an algorithm for measuring similarity between two sequences that vary in time or speed, which meets the characteristics of the EIs generated from FDPS. It is able to find an optimal match between two given sequences with certain restrictions. The sequences are "warped" non-linearly in the time dimension to determine a measure of their similarity independent of certain nonlinear variations in the time dimension. Let *s*1 and *s*<sup>2</sup> be two vectors with lengths of *m* and *n*. The goal of DTW is to find a mapping path {(*p*1, *q*1), (*p*2, *q*2), ⋯, (*pk* , *qk* )} such that the dis‐ tance on this mapping path ∑*<sup>i</sup>*=1 *<sup>k</sup>* <sup>|</sup>*s*1(*pi* )−*s*2(*qi* )| is minimal.

creases the computational complexity, the FDPS-based EIs shows a better effectiveness per‐ formance than TDPS. Secondly, the accuracy of R-wave detection affects the recognition performance of TDPS-based EIs heavily. For example, once a negative R-wave is detected or a positive R-wave is missed in one end, the EIs from true pairs will be dissimilar. Therefore, there would be a requirement for the given TDPS-based EI generation scheme that, the physiological signals shall be with an acceptable quality for peak detection. Though FDPSbased EI generation may also require a good signal quality, there is no evidence that the re‐ quirement is more constrict while compared to the TDPS-based one. Thirdly, from both the randomness performance and group similarity analysis, TDPS-based EIs shows a better per‐ formance than FDPS-based. In conclusion, TDPS-based EIs is superior in randomness and distinctiveness performance, while FDPS-based EIs is superior in effectiveness and robust‐

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

263

As presented in the workflow of the biometrics security in Section 1.3, the EIs can not only be used to identify individuals, but also be used to protect the transmission of keying mate‐ rials. The key distribution process in biometrics security model works as follows: one of the two sensors, called transmitter, hides the random symmetric key generated by its own using an EI obtained from the physiological signal. This hidden key is sent over to another sensor, called receiver, which uses its own version of EI to recover the random key after compensat‐ ing for the differences between its EI and the one used by the transmitter. The most common fuzzy methods used in biometrics security solution until now are fuzzy commitment and fuzzy vault, dependent on the characteristics of the EI generated. In this section, the two fuz‐ zy methods application will be detailed with a discussion of the specific fuzzy method to be

adopted for TDPS-based EIs and FDPS-based EIs according to their characteristics.

The block diagram of key distribution solution between communication parties based on the

cryptographic key need to be protected and its corresponding error-correction codeword, re‐ spectively, *EI* ∈{0,1}*<sup>n</sup>* represent the EI value at the transmitter used to protect keying materi‐

)=*h* (*K*), then the decommitment is successful and *K* ′

witness *EI*. It is obvious that the EI used in such a security model must be sequential and with the same length. In fact, in order to realize high-level security performance, *EI* used in

and *K*

at its own end, the receiver computes

⊕ *EI*)), where *f* is the relevant error-correction decoding

be a one-way hash function. The fuzzy commitment scheme is

is an incorrect witness that is not close enough to the original encrypting

^ <sup>⊕</sup> *EI*), where ⊕ is the bitwise XOR operation. To decommit

^ <sup>∈</sup>{0,1}*<sup>n</sup>* represent the

is the correct key *K*.

**3.1. Fuzzy commitment scheme applied in BSN security**

fuzzy commitment scheme is presented in Fig.9. Let *K* ∈{0,1}*<sup>k</sup>*

^ <sup>⊕</sup> (*EI* ′

ness performance.

**3. Key distribution solution**

als, and *h* :{0, 1}*<sup>n</sup>* →{0, 1}*<sup>l</sup>*

⊕ (*K*

^ , *EI*)=(*<sup>h</sup>* (*K*), *<sup>K</sup>*

^ , *EI*) using a witness *EI* ′

^ <sup>⊕</sup> *EI*))= *<sup>f</sup>* (*<sup>K</sup>*

defined as *F* (*K*

= *f* (*EI* ′

process. If *h* (*K* ′

Otherwise, *EI* ′

*F* (*K*

*K* ′

Fig.8 depicts the DTW distance distribution of EIs generated from the true pairs, i.e., two no‐ des on the same individual, with the SW and MWs methods on PPG data, respectively. It can be seen that with the SW EI generation scheme, 98% of the DTW distance between true pairs are less than 90, compared with 82% with the MWs, and thus exhibit a better perform‐ ance of group similarity than those with MWs.

**Figure 8.** DTW distance distribution of FDPS-based EIs of the true pairs.

#### *2.3.3. Performance comparison between two EI generation schemes*

In order to realize high recognition rate, the generated EIs should have the characteristics of effectiveness (being able to be generated fast and easily), robustness (resistance to uncertain‐ ty), randomness, distinctiveness (being similar for same subjects and differentiate for differ‐ ent subjects). Firstly, as about 30 seconds of ECG/PPG measurements are required to generate a 128-bit EI based on TDPS while only 5 seconds based on FDPS, and an R-wave detection is needed in TDPS-based EI generation scheme for the IPI measurement, which in‐ creases the computational complexity, the FDPS-based EIs shows a better effectiveness per‐ formance than TDPS. Secondly, the accuracy of R-wave detection affects the recognition performance of TDPS-based EIs heavily. For example, once a negative R-wave is detected or a positive R-wave is missed in one end, the EIs from true pairs will be dissimilar. Therefore, there would be a requirement for the given TDPS-based EI generation scheme that, the physiological signals shall be with an acceptable quality for peak detection. Though FDPSbased EI generation may also require a good signal quality, there is no evidence that the re‐ quirement is more constrict while compared to the TDPS-based one. Thirdly, from both the randomness performance and group similarity analysis, TDPS-based EIs shows a better per‐ formance than FDPS-based. In conclusion, TDPS-based EIs is superior in randomness and distinctiveness performance, while FDPS-based EIs is superior in effectiveness and robust‐ ness performance.

## **3. Key distribution solution**

Therefore, dynamic time warping (DTW) distance was selected to measure the group simi‐ larity between any pair of EIs generated from the same subject. DTW is an algorithm for measuring similarity between two sequences that vary in time or speed, which meets the characteristics of the EIs generated from FDPS. It is able to find an optimal match between two given sequences with certain restrictions. The sequences are "warped" non-linearly in the time dimension to determine a measure of their similarity independent of certain nonlinear variations in the time dimension. Let *s*1 and *s*<sup>2</sup> be two vectors with lengths of *m* and *n*. The goal of DTW is to find a mapping path {(*p*1, *q*1), (*p*2, *q*2), ⋯, (*pk* , *qk* )} such that the dis‐

*<sup>k</sup>* <sup>|</sup>*s*1(*pi*

)−*s*2(*qi*

Fig.8 depicts the DTW distance distribution of EIs generated from the true pairs, i.e., two no‐ des on the same individual, with the SW and MWs methods on PPG data, respectively. It can be seen that with the SW EI generation scheme, 98% of the DTW distance between true pairs are less than 90, compared with 82% with the MWs, and thus exhibit a better perform‐

[0,10) [20,30) [40,50) [60,70) [80,90) [100,120) [120,130) >=140

Range of DTW distance

In order to realize high recognition rate, the generated EIs should have the characteristics of effectiveness (being able to be generated fast and easily), robustness (resistance to uncertain‐ ty), randomness, distinctiveness (being similar for same subjects and differentiate for differ‐ ent subjects). Firstly, as about 30 seconds of ECG/PPG measurements are required to generate a 128-bit EI based on TDPS while only 5 seconds based on FDPS, and an R-wave detection is needed in TDPS-based EI generation scheme for the IPI measurement, which in‐

)| is minimal.

MWs Proposed SW

tance on this mapping path ∑*<sup>i</sup>*=1

262 New Trends and Developments in Biometrics

ance of group similarity than those with MWs.

0

**Figure 8.** DTW distance distribution of FDPS-based EIs of the true pairs.

*2.3.3. Performance comparison between two EI generation schemes*

0.05

0.1

0.15

Probability

0.2

0.25

0.3

0.35

As presented in the workflow of the biometrics security in Section 1.3, the EIs can not only be used to identify individuals, but also be used to protect the transmission of keying mate‐ rials. The key distribution process in biometrics security model works as follows: one of the two sensors, called transmitter, hides the random symmetric key generated by its own using an EI obtained from the physiological signal. This hidden key is sent over to another sensor, called receiver, which uses its own version of EI to recover the random key after compensat‐ ing for the differences between its EI and the one used by the transmitter. The most common fuzzy methods used in biometrics security solution until now are fuzzy commitment and fuzzy vault, dependent on the characteristics of the EI generated. In this section, the two fuz‐ zy methods application will be detailed with a discussion of the specific fuzzy method to be adopted for TDPS-based EIs and FDPS-based EIs according to their characteristics.

#### **3.1. Fuzzy commitment scheme applied in BSN security**

The block diagram of key distribution solution between communication parties based on the fuzzy commitment scheme is presented in Fig.9. Let *K* ∈{0,1}*<sup>k</sup>* and *K* ^ <sup>∈</sup>{0,1}*<sup>n</sup>* represent the cryptographic key need to be protected and its corresponding error-correction codeword, re‐ spectively, *EI* ∈{0,1}*<sup>n</sup>* represent the EI value at the transmitter used to protect keying materi‐ als, and *h* :{0, 1}*<sup>n</sup>* →{0, 1}*<sup>l</sup>* be a one-way hash function. The fuzzy commitment scheme is defined as *F* (*K* ^ , *EI*)=(*<sup>h</sup>* (*K*), *<sup>K</sup>* ^ <sup>⊕</sup> *EI*), where ⊕ is the bitwise XOR operation. To decommit *F* (*K* ^ , *EI*) using a witness *EI* ′ at its own end, the receiver computes *K* ′ = *f* (*EI* ′ ⊕ (*K* ^ <sup>⊕</sup> *EI*))= *<sup>f</sup>* (*<sup>K</sup>* ^ <sup>⊕</sup> (*EI* ′ ⊕ *EI*)), where *f* is the relevant error-correction decoding process. If *h* (*K* ′ )=*h* (*K*), then the decommitment is successful and *K* ′ is the correct key *K*. Otherwise, *EI* ′ is an incorrect witness that is not close enough to the original encrypting witness *EI*. It is obvious that the EI used in such a security model must be sequential and with the same length. In fact, in order to realize high-level security performance, *EI* used in fuzzy commitment must have the performance of distinctiveness and time-variance to en‐ sure the invulnerability.

matchings in the fuzzy vault. All the candidate points are identified together with their pair values in the vault to form a set *S*. Let *U* denotes the number of pairs in *S*. To reconstruct the polynomial with *m* degree, all possible combinations of *m* + 1 points are identified, with

the polynomial using Lagrange interpolating technique. The coefficients in the generated polynomial is mapped back and concatenated in the same order as encoding to generate an

**Sensor 1** *K* **Sensor 2**

**Set A Set B**

**Polynomial construct**

) combinations. Each of the possible combinations is used to recover

Physiological Signal Based Biometrics for Securing Body Sensor Network

**Receiving Vault**

**matching**

**Biometric identifier**

a1 a3

http://dx.doi.org/10.5772/51856

265

<sup>0</sup> <sup>2</sup> <sup>4</sup> <sup>6</sup> <sup>8</sup> <sup>10</sup> <sup>12</sup> <sup>14</sup> <sup>16</sup> <sup>18</sup> <sup>20</sup> <sup>0</sup>

a1 a3 a5 a4 a2

fail

5 10 15

P(x)

P(x)

a2 a5 a4

x

x

**Reconstruct Polynomial**

*K*

*MAC K MAC K* () ( ) =

*K*

*<sup>Y</sup> <sup>N</sup>*

. The cryptographic key *K* can be retrieved while the message authentication

equals to *MAC*(*K*) if the two EIs generated at both ends are with high similarity.

a total number of (

a1 a3

<sup>0</sup> <sup>2</sup> <sup>4</sup> <sup>6</sup> <sup>8</sup> <sup>10</sup> <sup>12</sup> <sup>14</sup> <sup>16</sup> <sup>18</sup> <sup>20</sup> <sup>0</sup>

<sup>0</sup> <sup>2</sup> <sup>4</sup> <sup>6</sup> <sup>8</sup> <sup>10</sup> <sup>12</sup> <sup>14</sup> <sup>16</sup> <sup>18</sup> <sup>20</sup> <sup>0</sup>

a1 a3 a5 a4 a2

a1 a3 a5 a4 a2

5 10 15

P(x)

5 10 15

P(x)

P(x)

a2 a5 a4

n-bit code *K* ′

code of *K* ′

*U m* + 1

**Projection**

**Biometric identifier**

**Adding chaff points**

> **Sending Vault**

**3.3. Performance comparison between different application scenarios**

In order to realize high recognition performance and security, a suitable key distribution sol‐ ution should be selected based on the specific characteristics of EIs generated based on TDPS and FDPS. In this section, we conduct a series of experiments to evaluate the perform‐ ance of different key distribution solutions. Firstly, a detailed recognition performance in terms of False Accept Rate (FAR)/False Reject Rate (FRR) is conducted for different EI gener‐ ation scheme with different key distribution solution to demonstrate the suitable key distri‐ bution solution for different EIs generated. Then, the security performance of different fuzzy

x

x

x

**Figure 10.** Key distribution solution based on fuzzy vault scheme

**Figure 9.** Key distribution solution based on fuzzy commitment scheme

#### **3.2. Fuzzy vault scheme applied in BSN**

Fig.10 gives the block diagram of key distribution solution between communication parties based on the fuzzy vault scheme applied in BSN security. In fuzzy vault based key distribu‐ tion scheme, let *K* ∈{0,1}*<sup>n</sup>* represent the cryptographic key need to be protected, *ai* <sup>∈</sup>{0, 1}*<sup>k</sup>* , *i* =1⋯*M* represent the binary biometric features derived from the generated EI used to protect keying materials. A polynomial *P*(*x*)=*cmx <sup>m</sup>* <sup>+</sup> *cm*−1*<sup>x</sup> <sup>m</sup>*−<sup>1</sup> <sup>+</sup> *cm*−2*<sup>x</sup> <sup>m</sup>*−<sup>2</sup> <sup>+</sup> <sup>⋯</sup>*c*1*<sup>x</sup>* <sup>+</sup> *<sup>c</sup>*<sup>0</sup> is created for binding of *K* and *ai* <sup>∈</sup>{0, 1}*<sup>k</sup>* , *i* =1⋯*M* by segmenting *K* as its coefficients with the form of *K* =*cm* | |*cm*−<sup>1</sup> | |*cm*−<sup>2</sup> | | ⋯ | |*c*0, where *m* is the degree of the polynomial. The polynomial *P*(*x*) is then evaluated on each of the feature points *Xi* , where *Xi* is an integer number corresponds to binary feature *ai* . The generated pairs {(*Xi* , *P*(*Xi* )), *i* =1⋯*M* } are termed the genuine set *G*. Then the transmitter generates the chaff points set *C* ={(*uj* , *vj* ), *j* =1⋯ *Nc*}, where *Nc*≫*M* , *uj* ≠ *Xi* and each pair does not lie on the polynomial, i.e. *vj* ≠ *f* (*uj* ). The final vault is constructed by taking the union of the two sets, i.e. *G* ∪*C*, combined with the message authentication code (e.g. MD5, SHA1)of *K*, denoted as *MAC*(*K*), and pass through a scrambler so that it is not clear which are the feature points and which are the chaff points. The receiver decodes the fuzzy vault using binary biometric features *a* ′ *<sup>i</sup>* ∈{0, 1}*<sup>k</sup>* , *i* =1⋯*M* derived from the EI generated by itself by searching for the matchings in the fuzzy vault. All the candidate points are identified together with their pair values in the vault to form a set *S*. Let *U* denotes the number of pairs in *S*. To reconstruct the polynomial with *m* degree, all possible combinations of *m* + 1 points are identified, with a total number of ( *U m* + 1 ) combinations. Each of the possible combinations is used to recover the polynomial using Lagrange interpolating technique. The coefficients in the generated polynomial is mapped back and concatenated in the same order as encoding to generate an n-bit code *K* ′ . The cryptographic key *K* can be retrieved while the message authentication code of *K* ′ equals to *MAC*(*K*) if the two EIs generated at both ends are with high similarity.

**Figure 10.** Key distribution solution based on fuzzy vault scheme

fuzzy commitment must have the performance of distinctiveness and time-variance to en‐

Synchronization

Receiver

EI generation scheme

> *K*ˆ Error-Correction decoding *K*

*K EI* ˆ

*h K*( )

Fig.10 gives the block diagram of key distribution solution between communication parties based on the fuzzy vault scheme applied in BSN security. In fuzzy vault based key distribu‐ tion scheme, let *K* ∈{0,1}*<sup>n</sup>* represent the cryptographic key need to be protected,

used to protect keying materials. A polynomial *P*(*x*)=*cmx <sup>m</sup>* <sup>+</sup> *cm*−1*<sup>x</sup> <sup>m</sup>*−<sup>1</sup> <sup>+</sup> *cm*−2*<sup>x</sup> <sup>m</sup>*−<sup>2</sup> <sup>+</sup> <sup>⋯</sup>*c*1*<sup>x</sup>* <sup>+</sup> *<sup>c</sup>*<sup>0</sup>

the form of *K* =*cm* | |*cm*−<sup>1</sup> | |*cm*−<sup>2</sup> | | ⋯ | |*c*0, where *m* is the degree of the polynomial. The

termed the genuine set *G*. Then the transmitter generates the chaff points set

combined with the message authentication code (e.g. MD5, SHA1)of *K*, denoted as *MAC*(*K*), and pass through a scrambler so that it is not clear which are the feature points and which are the chaff points. The receiver decodes the fuzzy vault using binary biometric

polynomial *P*(*x*) is then evaluated on each of the feature points *Xi*

, *i* =1⋯*M* represent the binary biometric features derived from the generated EI

. The generated pairs {(*Xi*

, *i* =1⋯*M* derived from the EI generated by itself by searching for the

). The final vault is constructed by taking the union of the two sets, i.e. *G* ∪*C*,

*EI'*

Physiological Signal

*K*

, *i* =1⋯*M* by segmenting *K* as its coefficients with

, where *Xi* is an integer

)), *i* =1⋯*M* } are

, *P*(*Xi*

and each pair does not lie on the polynomial,

*hK hK* () ( ) =

*<sup>N</sup>* fail

*Y*

EI generation scheme

*EI*

ˆ *F K EI* (, )

ˆ ( ( ), ) *F K EI h K K EI* = 

Physiological Signal

*<sup>K</sup>*<sup>ˆ</sup> <sup>ˆ</sup> (, )

**Figure 9.** Key distribution solution based on fuzzy commitment scheme

*K*

Error-Correction encoding

**3.2. Fuzzy vault scheme applied in BSN**

is created for binding of *K* and *ai* <sup>∈</sup>{0, 1}*<sup>k</sup>*

number corresponds to binary feature *ai*

), *j* =1⋯ *Nc*}, where *Nc*≫*M* , *uj* ≠ *Xi*

*ai* <sup>∈</sup>{0, 1}*<sup>k</sup>*

*C* ={(*uj*

i.e. *vj* ≠ *f* (*uj*

features *a* ′

*<sup>i</sup>* ∈{0, 1}*<sup>k</sup>*

, *vj*

Transmitter

sure the invulnerability.

264 New Trends and Developments in Biometrics

#### **3.3. Performance comparison between different application scenarios**

In order to realize high recognition performance and security, a suitable key distribution sol‐ ution should be selected based on the specific characteristics of EIs generated based on TDPS and FDPS. In this section, we conduct a series of experiments to evaluate the perform‐ ance of different key distribution solutions. Firstly, a detailed recognition performance in terms of False Accept Rate (FAR)/False Reject Rate (FRR) is conducted for different EI gener‐ ation scheme with different key distribution solution to demonstrate the suitable key distri‐ bution solution for different EIs generated. Then, the security performance of different fuzzy methods are presented. At last, the computational complexity performances are conducted for different key distribution solutions with the appropriate EI generation scheme.

fined degree of polynomial. It can be seen that the fuzzy commitment scheme shows a better recognition performance with a minumum HTER of less than 1.46% and 3.19% on 14 and 85 subjects, compared to 3.4% and 5.6% with fuzzy vault scheme. The results indicate that the fuzzy commitment scheme is superior to fuzzy vault scheme in different settings, such as

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

267

As presented in Section 2.3.2, the two feature sets generated based on FDPS from the trans‐ mitter and receiver are with different numbers of points and it is common to have matching points at different orders of the two sets, thus fuzzy commitment scheme is not suitable for FDPS-based EIs due to its requirement for correspondence of features in terms of order. Therefore, fuzzy vault scheme is probably the only approriate solution to protect the trans‐ mmission of keying materials with FDPS-based EIs. The data we used for performance eval‐ uation include ECG data of 20 subjects from the physioBank database (http:// www.physionet.org/physiobank), including 10 healthy people and 10 people with different kinds of diseases, which were simultaneously collected from two leads on each subject at a sampling rate of 360 Hz, and two-channel PPG data at a sampling rate of 1000 Hz from 14 subjects in Exp.I Fig.12 depicts the FAR/FRR curves with fuzzy vault scheme. It can be seen that fuzzy vault shows an recognition performance with a minumum HTER of 5.2% and

0 5 10 15 20 25 30 35 40 45 50

FRR(ECG) FAR(ECG) FRR(PPG) FAR(PPG)

Degree of Polynomial

From the above analysis we can see, fuzzy commitment scheme is suitable for TDPS-based EIs while fuzzy vault scheme for FDPS-based EIs. In addition, TDPS-based solution shows a

the lab and the clinical setting, for TDPS-based EIs.

8.9% on ECG and PPG, respectively.

Probability

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

**Figure 12.** FAR/FRR curves with FDPS-based EIs based on fuzzy vault scheme

better recognition rate performance than FDPS-based one.

#### *3.3.1. FAR/FRR performance*

FAR and FRR are two important indexes to evaluate the recognition rate performance of a biometric system, where FAR is the probability that a system incorrectly matches the input pattern from false pairs, FRR is the probability that a system fails to detect a match between the true pairs. The most suitable fuzzy method for different EIs should achieve a minumum half total error rate (HTER) that equals (FAR+FRR)/2.

**Figure 11.** FAR/FRR curves with TDPS-based EIs. (a) Fuzzy commitment scheme; (b) Fuzzy vault scheme

For TDPS-based EIs, we conduct our performance evaluation based on data from two ex‐ periments described in Section 2.3. In Exp.I, i.e. there were in total 14 healthy subjects and three channels of physiological signals (including 1-channel ECG and 2-channel PPG) with a duration of 2-3 minutes were simultaneously recorded from each subject. In Exp.II, there were in total 85 clinical subjects and two channels of physiological signals (including onechannel ECG and one-channel PPG) with a duration of 40 seconds were simultaneously re‐ corded from each subject on three or four days within a two-month period. Data from the both experiments are with 12-bit A/D resolution and at a sampling rate of 1000 Hz. Fig.11(a) depicts the FAR/FRR curves with fuzzy commitment scheme, where FRR is the rate at which the two EIs generated from the same subject during the same period are unmatched, i.e. Hamming distance is larger than a specific threshold, FAR is the rate at which the two EIs generated from the different subjects or during different periods are matched, i.e. Hamming distance is larger than a specific threshold. Fig.11(b) depicts the FAR/FRR curves with fuzzy vault scheme, where FRR is the rate at which the two EIs generated from the same subject during the same period are unmatched, i.e. matching number is smaller than a predefined degree of polynomial, FAR is the rate at which the two EIs generated from the different sub‐ jects or during different periods are matched, i.e. matching number is smaller than a prede‐ fined degree of polynomial. It can be seen that the fuzzy commitment scheme shows a better recognition performance with a minumum HTER of less than 1.46% and 3.19% on 14 and 85 subjects, compared to 3.4% and 5.6% with fuzzy vault scheme. The results indicate that the fuzzy commitment scheme is superior to fuzzy vault scheme in different settings, such as the lab and the clinical setting, for TDPS-based EIs.

methods are presented. At last, the computational complexity performances are conducted

FAR and FRR are two important indexes to evaluate the recognition rate performance of a biometric system, where FAR is the probability that a system incorrectly matches the input pattern from false pairs, FRR is the probability that a system fails to detect a match between the true pairs. The most suitable fuzzy method for different EIs should achieve a minumum

for different key distribution solutions with the appropriate EI generation scheme.

**Figure 11.** FAR/FRR curves with TDPS-based EIs. (a) Fuzzy commitment scheme; (b) Fuzzy vault scheme

For TDPS-based EIs, we conduct our performance evaluation based on data from two ex‐ periments described in Section 2.3. In Exp.I, i.e. there were in total 14 healthy subjects and three channels of physiological signals (including 1-channel ECG and 2-channel PPG) with a duration of 2-3 minutes were simultaneously recorded from each subject. In Exp.II, there were in total 85 clinical subjects and two channels of physiological signals (including onechannel ECG and one-channel PPG) with a duration of 40 seconds were simultaneously re‐ corded from each subject on three or four days within a two-month period. Data from the both experiments are with 12-bit A/D resolution and at a sampling rate of 1000 Hz. Fig.11(a) depicts the FAR/FRR curves with fuzzy commitment scheme, where FRR is the rate at which the two EIs generated from the same subject during the same period are unmatched, i.e. Hamming distance is larger than a specific threshold, FAR is the rate at which the two EIs generated from the different subjects or during different periods are matched, i.e. Hamming distance is larger than a specific threshold. Fig.11(b) depicts the FAR/FRR curves with fuzzy vault scheme, where FRR is the rate at which the two EIs generated from the same subject during the same period are unmatched, i.e. matching number is smaller than a predefined degree of polynomial, FAR is the rate at which the two EIs generated from the different sub‐ jects or during different periods are matched, i.e. matching number is smaller than a prede‐

*3.3.1. FAR/FRR performance*

266 New Trends and Developments in Biometrics

half total error rate (HTER) that equals (FAR+FRR)/2.

As presented in Section 2.3.2, the two feature sets generated based on FDPS from the trans‐ mitter and receiver are with different numbers of points and it is common to have matching points at different orders of the two sets, thus fuzzy commitment scheme is not suitable for FDPS-based EIs due to its requirement for correspondence of features in terms of order. Therefore, fuzzy vault scheme is probably the only approriate solution to protect the trans‐ mmission of keying materials with FDPS-based EIs. The data we used for performance eval‐ uation include ECG data of 20 subjects from the physioBank database (http:// www.physionet.org/physiobank), including 10 healthy people and 10 people with different kinds of diseases, which were simultaneously collected from two leads on each subject at a sampling rate of 360 Hz, and two-channel PPG data at a sampling rate of 1000 Hz from 14 subjects in Exp.I Fig.12 depicts the FAR/FRR curves with fuzzy vault scheme. It can be seen that fuzzy vault shows an recognition performance with a minumum HTER of 5.2% and 8.9% on ECG and PPG, respectively.

**Figure 12.** FAR/FRR curves with FDPS-based EIs based on fuzzy vault scheme

From the above analysis we can see, fuzzy commitment scheme is suitable for TDPS-based EIs while fuzzy vault scheme for FDPS-based EIs. In addition, TDPS-based solution shows a better recognition rate performance than FDPS-based one.

## *3.3.2. Security analysis*

Suppose the EIs generated are random enough, the security issues in the proposed key dis‐ tribution solution primarily exist during its package exchange process by brute-forcing the key (*K*) or EI directly. Therefore, to ensure the security of the key distribution protocol, the information contained in EI must be larger than that in *K*.

For fuzzy commitment scheme, an eavesdropper can try out each bit of *K* by brute-force at‐ tack. Also, he can try out most of bits in EI to reconstruct the same *K*. Suppose the length of *K* is *l*, the computation requirement of directly attack on *K* in terms of its equivalence to brute-forcing a key of a particular length (bits) is *l*. The number of attempts by attacking EI depends on the length of *K* and the ability of its corresponding error-correction. Take Reed-Solomon as the error-correction code for example, a redundancy code of 2×*t* bits should be attached to correct *t*-bit errors, thus the length of EI should be equal to*l* + 2×*t*. As the errorcorrection code can correct *t*-bit errors, an attempt of *l* + 2×*t* −*t* =*l* + *t* bits can reconstruct *K* successfully. In conclusion, the computation requirement in terms of its equivalence to brute-forcing a key of a particular length (bits) is min(*l*, *l* + *t*)=*l*. In another word, the securi‐ ty of fuzzy commitment depends on the security of *K* directly.

16 17 18 19 20 21 22 23 24 25

Vault Size=200 Vault Size=300 Vault Size=400 Upper Limit

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

269

), where *N* is the number of IPIs

^ , *EI*) and *EI* ′

in the receiv‐

m+1

The time complexity involved in TDPS-based key distribution solution requires the follow‐ ing tasks: 1)R-wave detection; 2) IPI-based EI generation process; 3) Error-correction encod‐ ing process; 4) Error-correction decoding process. From the FAR/FRR performance analysis we can see, the Hamming distance to achieve minimum HTER is about 18. Thus, from the security analysis, in order to ensure the security of 128 bits, an EI of 164 bits should be gen‐ erated. Take *q* =4 for example, 41 IPIs should be calculated, which means 42 R-waves should be detected. Take difference threshold algorithm for example, which has the minimum com‐ putational complexity in R-wave detection algorithms, the time complexity of R-wave detec‐ tion process on *n* points is *O*(*n*). It was demonstrated by Bao et al (2009) that the time

used. Take Reed-Solomon as the error-correction code, the time complexity for the encoding process is *O*(*N* .*q*), and for the decoding process based on Berlekamp–Massey algorithm the

required for implementing proposed schemes. Excluding the dynamic occupied memory due to R-wave detection process and error-correction encoding/decoding process, the pri‐

The time complexity involved in FDPS-based key distribution solution requires the follow‐ ing tasks: 1) FFT computation; 2) Peak detection; 3) EI generation; 4) Key hiding (polynomial evaluation); 5) Key un-hiding (Lagrange interpolation). For the FFT computation process performed on *n* points, the time complexity is *O*(*n*log*n*). As *M* points of FFT coefficients that are selected to perform a peak detection process, the time complexity for peak detection is

). As *q* is a fixed number, the time complexity of proposed solu‐

^ , *EI*) while *<sup>F</sup>* (*<sup>K</sup>*

). The space complexity is estimated in terms of memory

70

complexity for IPI-based EI generation process is *O*(*N* <sup>2</sup>

2

mary static components in the transmitter is *F* (*K*

er, and the overall memory required is 84 bytes.

**Figure 13.** Security of fuzzy vault scheme

time complexity is *O*((*N* .*q*)

tion can be expressed as *O*(*n* + *N* <sup>2</sup>

80

90

100

Security of Vault(Bits)

110

120

130

For fuzzy vault scheme, except for brute-forcing *K* directly, an eavesdropper can record the vault and try to construct the hidden polynomial from it. As described above, the computa‐ tion requirement of directly attack on *K* is *l*. Suppose the degree of the polynomial is *m*, as the feature points are hidden among a much larger number of chaff points, whose values are randomly distributed in the same range in some situation, an adversary is able to try out each group of *m* + 1 points in the vault to get the correct polynomial, the average attempts are (*<sup>L</sup> m* + 1 ) / 2, where *L* is the vault size and *<sup>L</sup>* <sup>≤</sup>2*<sup>p</sup>* . Thus, the security of vault is a balance act between the vault size *L* and the degree of the polynomial *m*, but subject to *p* and *l*. In conclusion, the computation requirement in terms of its equivalence to brute-forcing a key (*L*

of a particular length (bits) is min(log2 *m*+1 ) , *l*). The security of the vault for different values of *m* and different number of vault size in the condition *l* =128 and *p* =9 is presented in Fig.13. For ease of understanding, we represent this computation requirement in terms of its equiv‐ alence to brute-forcing a key of a particular length (bits). As expected, increasing the number of chaff points increases the security provided by the vault, but the security is subject to *p* and *l*. Higher the order of the polynomial means higher security, where more common fea‐ ture points shall be hold by two ends.

#### *3.3.3. Computational complexity*

As described in Section 3.3.1, the suitable fuzzy method for TDPS-based EIs is fuzzy com‐ mitment while fuzzy vault for FDPS-based EIs. We estimate the cost performance of pro‐ posed key distribution solutions in terms of computational complexity, including time complexity and space complexity.

**Figure 13.** Security of fuzzy vault scheme

*3.3.2. Security analysis*

268 New Trends and Developments in Biometrics

are (*<sup>L</sup>*

*m* + 1

Suppose the EIs generated are random enough, the security issues in the proposed key dis‐ tribution solution primarily exist during its package exchange process by brute-forcing the key (*K*) or EI directly. Therefore, to ensure the security of the key distribution protocol, the

For fuzzy commitment scheme, an eavesdropper can try out each bit of *K* by brute-force at‐ tack. Also, he can try out most of bits in EI to reconstruct the same *K*. Suppose the length of *K* is *l*, the computation requirement of directly attack on *K* in terms of its equivalence to brute-forcing a key of a particular length (bits) is *l*. The number of attempts by attacking EI depends on the length of *K* and the ability of its corresponding error-correction. Take Reed-Solomon as the error-correction code for example, a redundancy code of 2×*t* bits should be attached to correct *t*-bit errors, thus the length of EI should be equal to*l* + 2×*t*. As the errorcorrection code can correct *t*-bit errors, an attempt of *l* + 2×*t* −*t* =*l* + *t* bits can reconstruct *K* successfully. In conclusion, the computation requirement in terms of its equivalence to brute-forcing a key of a particular length (bits) is min(*l*, *l* + *t*)=*l*. In another word, the securi‐

For fuzzy vault scheme, except for brute-forcing *K* directly, an eavesdropper can record the vault and try to construct the hidden polynomial from it. As described above, the computa‐ tion requirement of directly attack on *K* is *l*. Suppose the degree of the polynomial is *m*, as the feature points are hidden among a much larger number of chaff points, whose values are randomly distributed in the same range in some situation, an adversary is able to try out each group of *m* + 1 points in the vault to get the correct polynomial, the average attempts

act between the vault size *L* and the degree of the polynomial *m*, but subject to *p* and *l*. In conclusion, the computation requirement in terms of its equivalence to brute-forcing a key

*m* and different number of vault size in the condition *l* =128 and *p* =9 is presented in Fig.13. For ease of understanding, we represent this computation requirement in terms of its equiv‐ alence to brute-forcing a key of a particular length (bits). As expected, increasing the number of chaff points increases the security provided by the vault, but the security is subject to *p* and *l*. Higher the order of the polynomial means higher security, where more common fea‐

As described in Section 3.3.1, the suitable fuzzy method for TDPS-based EIs is fuzzy com‐ mitment while fuzzy vault for FDPS-based EIs. We estimate the cost performance of pro‐ posed key distribution solutions in terms of computational complexity, including time

(*L m*+1 ) . Thus, the security of vault is a balance

, *l*). The security of the vault for different values of

information contained in EI must be larger than that in *K*.

ty of fuzzy commitment depends on the security of *K* directly.

) / 2, where *L* is the vault size and *<sup>L</sup>* <sup>≤</sup>2*<sup>p</sup>*

of a particular length (bits) is min(log2

ture points shall be hold by two ends.

*3.3.3. Computational complexity*

complexity and space complexity.

The time complexity involved in TDPS-based key distribution solution requires the follow‐ ing tasks: 1)R-wave detection; 2) IPI-based EI generation process; 3) Error-correction encod‐ ing process; 4) Error-correction decoding process. From the FAR/FRR performance analysis we can see, the Hamming distance to achieve minimum HTER is about 18. Thus, from the security analysis, in order to ensure the security of 128 bits, an EI of 164 bits should be gen‐ erated. Take *q* =4 for example, 41 IPIs should be calculated, which means 42 R-waves should be detected. Take difference threshold algorithm for example, which has the minimum com‐ putational complexity in R-wave detection algorithms, the time complexity of R-wave detec‐ tion process on *n* points is *O*(*n*). It was demonstrated by Bao et al (2009) that the time complexity for IPI-based EI generation process is *O*(*N* <sup>2</sup> ), where *N* is the number of IPIs used. Take Reed-Solomon as the error-correction code, the time complexity for the encoding process is *O*(*N* .*q*), and for the decoding process based on Berlekamp–Massey algorithm the time complexity is *O*((*N* .*q*) 2 ). As *q* is a fixed number, the time complexity of proposed solu‐ tion can be expressed as *O*(*n* + *N* <sup>2</sup> ). The space complexity is estimated in terms of memory required for implementing proposed schemes. Excluding the dynamic occupied memory due to R-wave detection process and error-correction encoding/decoding process, the pri‐ mary static components in the transmitter is *F* (*K* ^ , *EI*) while *<sup>F</sup>* (*<sup>K</sup>* ^ , *EI*) and *EI* ′ in the receiv‐ er, and the overall memory required is 84 bytes.

The time complexity involved in FDPS-based key distribution solution requires the follow‐ ing tasks: 1) FFT computation; 2) Peak detection; 3) EI generation; 4) Key hiding (polynomial evaluation); 5) Key un-hiding (Lagrange interpolation). For the FFT computation process performed on *n* points, the time complexity is *O*(*n*log*n*). As *M* points of FFT coefficients that are selected to perform a peak detection process, the time complexity for peak detection is *M* , where *M* =150 in our experiment. EI generation scheme includes an addition operation and a modulo operation on each feature point. The number of feature points depends on peak indexes detected, the time complexity of EI generation process is *O*(*βM* ), where *β* is the rate between peaks detected and the FFT coefficients selected and thus 0<*β* <1. The poly‐ nomial evaluation in key hiding process would require 48×*m*(*m* + 1)/ 2 operations, so the time complexity of key hiding process is 48×*m*(*m* + 1)/ 2. It is demonstrated by J.P. Berrut that the improved Lagrange interpolation, i.e., Barycentric interpolation, requires only *O*(*m*) operations as opposed to *O*(*m*<sup>2</sup> ) for evaluating the Lagrange basis individually. Therefore,

**Computational complexity Task Value**

FFT computation *O*(*n*log*n*)

Physiological Signal Based Biometrics for Securing Body Sensor Network

Peak detection *M* =150

Randomized process *O*(β*M* )

Chaff points (i.e. 400) 2.05KB

Feature points (i.e. 48) 252B

Feature points (i.e. 48) 252B

Vault (i.e. 448) 2.302KB

Key un-hiding (Polynomial

reconstruction)

Key hiding 48×*m*(*m* + 1) / 2

*O*(( *N*2 *m* + 1

http://dx.doi.org/10.5772/51856

)*m*)=*O*(*m*)

271

Transmitter *O*(*n*log*n*)

Receiver *O*(*m*)

Transmitter (2.302KB)

Receiver (2.552KB)

In the biometrics solution for BSN security, physiological signals within human body are used to generate dynamic EIs, which is not only used to realize node identification, but also protect the transmission of keying materials. In this chapter, the procedures of biometric sol‐ utions for securing BSN, including the EI generation scheme and relevant key distribution solution, have been described. From the experimental results we can see that, TDPS-based EI generation scheme is superior in randomness and recognition performance, while FDPSbased scheme has advantage on its real-time performance and robustness. The two common used fuzzy methods, including fuzzy commitment scheme and fuzzy vault scheme, also have their own advantages and disadvantages. Fuzzy commitment can achieve low compu‐ tation complexity and low memory occupied, but it is not suitable for EIs that are un-or‐ dered or with different length. Fuzzy vault scheme can be suitable to most of cases, but with a high computation complexity and memory occupied. To realize high recognition perform‐ ance, fuzzy commitment should be selected for TDPS-based EIs, called TDPS-based solu‐ tion, while fuzzy vault for FDPS-based EIs, called FDPS-based solution. There are a lot of

issues need to be further studied to make it applicable into practical BSN platforms.

The challenges of TDPS-based solution primary exist in the EI generation process, where a sig‐ nal of about 30s is needed to generate a 128-bit EI. Firstly, how to increase the positive detec‐ tion rate of R-wave with lower computational complexity or design a more robust EI generation scheme being little influenced by the precision of R-wave should be studied to in‐

FDPS-based with fuzzy vault

**Table 2.** Computational complexity

**4. Conclusion**

Time complexity *O*(*n*log*n*) Static space complexity (4.854KB)

the time complexity of key un-hiding process is reduced to *O*((*N*<sup>2</sup> *m* + 1 )*m*)=*O*(*m*), where *N*<sup>2</sup> is

the number of feature points generated at the receiver. As *m* and *M* are fixed numbers, the time complexity of proposed solution can be expressed as *O*(*n*log*n*). From the FAR/FRR per‐ formance analysis we can see, the degree of polynomial to achieve minimum HTER is about 23. Thus, in order to realize the security of 128 bits, the vault size should be larger than 400. Excluding the dynamic occupied memory due to FFT process and randomized process, the primary static components of the memory required are the physiological features (9 bit val‐ ues, about 48 for PPG and ECG for example) and their polynomial projects (12 bit values), chaff points (400 for example to realize the security of 128 bits, 9 bit x-values and 12 bit yvalues). The overall memory required is 4.854KB.

Table 2 gives the detailed computational complexity of the TDPS-based and FDPS-based key distribution solution, separately. It can be seen that TDPS-based key distribution solution is superior in space complexity with only 84B of memory required, compared to 4.854KB for FDPS-based solution.



**Table 2.** Computational complexity

## **4. Conclusion**

*M* , where *M* =150 in our experiment. EI generation scheme includes an addition operation and a modulo operation on each feature point. The number of feature points depends on peak indexes detected, the time complexity of EI generation process is *O*(*βM* ), where *β* is the rate between peaks detected and the FFT coefficients selected and thus 0<*β* <1. The poly‐ nomial evaluation in key hiding process would require 48×*m*(*m* + 1)/ 2 operations, so the time complexity of key hiding process is 48×*m*(*m* + 1)/ 2. It is demonstrated by J.P. Berrut that the improved Lagrange interpolation, i.e., Barycentric interpolation, requires only *O*(*m*)

the number of feature points generated at the receiver. As *m* and *M* are fixed numbers, the time complexity of proposed solution can be expressed as *O*(*n*log*n*). From the FAR/FRR per‐ formance analysis we can see, the degree of polynomial to achieve minimum HTER is about 23. Thus, in order to realize the security of 128 bits, the vault size should be larger than 400. Excluding the dynamic occupied memory due to FFT process and randomized process, the primary static components of the memory required are the physiological features (9 bit val‐ ues, about 48 for PPG and ECG for example) and their polynomial projects (12 bit values), chaff points (400 for example to realize the security of 128 bits, 9 bit x-values and 12 bit y-

Table 2 gives the detailed computational complexity of the TDPS-based and FDPS-based key distribution solution, separately. It can be seen that TDPS-based key distribution solution is superior in space complexity with only 84B of memory required, compared to 4.854KB for

**Computational complexity Task Value**

process

*K*

*F* (*K*

Transmitter *O*(*n* + *N* <sup>2</sup> )

Receiver *O*(*N* <sup>2</sup> )

Transmitter (34B)

Receiver (50B)

) for evaluating the Lagrange basis individually. Therefore,

*m* + 1

R-wave detection *O*(*n*)

Error correction encoding *O*(*N* .*q*)

Error correction decoding *O*((*N* .*q*)

^ <sup>⊕</sup> *EI* 18B

hash(K) 16B

^, *EI*) 34B

*E I* ′ 16B

*O*(*N* <sup>2</sup> )

> 2 )

IPI\_based EI generation

)*m*)=*O*(*m*), where *N*<sup>2</sup> is

operations as opposed to *O*(*m*<sup>2</sup>

270 New Trends and Developments in Biometrics

FDPS-based solution.

TDPS-based with

fuzzy commitment

the time complexity of key un-hiding process is reduced to *O*((*N*<sup>2</sup>

values). The overall memory required is 4.854KB.

Time complexity *O*(*n* + *N* <sup>2</sup> )

Static space complexity (84B) In the biometrics solution for BSN security, physiological signals within human body are used to generate dynamic EIs, which is not only used to realize node identification, but also protect the transmission of keying materials. In this chapter, the procedures of biometric sol‐ utions for securing BSN, including the EI generation scheme and relevant key distribution solution, have been described. From the experimental results we can see that, TDPS-based EI generation scheme is superior in randomness and recognition performance, while FDPSbased scheme has advantage on its real-time performance and robustness. The two common used fuzzy methods, including fuzzy commitment scheme and fuzzy vault scheme, also have their own advantages and disadvantages. Fuzzy commitment can achieve low compu‐ tation complexity and low memory occupied, but it is not suitable for EIs that are un-or‐ dered or with different length. Fuzzy vault scheme can be suitable to most of cases, but with a high computation complexity and memory occupied. To realize high recognition perform‐ ance, fuzzy commitment should be selected for TDPS-based EIs, called TDPS-based solu‐ tion, while fuzzy vault for FDPS-based EIs, called FDPS-based solution. There are a lot of issues need to be further studied to make it applicable into practical BSN platforms.

The challenges of TDPS-based solution primary exist in the EI generation process, where a sig‐ nal of about 30s is needed to generate a 128-bit EI. Firstly, how to increase the positive detec‐ tion rate of R-wave with lower computational complexity or design a more robust EI generation scheme being little influenced by the precision of R-wave should be studied to in‐ crease the robustness performance of the solution. Secondly, a faster EI generation scheme based on minimum number of IPIs should be addressed to increase its real-time performance.

[5] D. Liu, P. Ning (2003). *Establishing pariwise keys in distributed sensor networks,* proceed‐ ings of the 10th ACM Conference on Computer and Communication, pp. 42-51.

Physiological Signal Based Biometrics for Securing Body Sensor Network

http://dx.doi.org/10.5772/51856

273

[6] S. Cherukuri, K. K. Venkatasubramanian, S. K. S. Gupta (2003). *BioSec: a biometric based approach for securing communication in wireless networks of biosensors implanted in the human body,* Proc. IEEE International Conference Parallel Processing Workshop,

[7] S. D. Bao, Y. T. Zhang, and L. F. Shen (2005). *Physiological Signal Based Entity Authenti‐ cation for Body Area Sensor Networks and Mobile Healthcare Systems,* Proc. 27th IEEE

[8] S. D. Bao, C. C. Y. Poon, Y. T. Zhang, and L. F. Shen (2008). *Using the timing informa‐ tion of heartbeats as an entity identifier to secure body sensor network,* IEEE transactions on

[9] Fengyuan Xu, Zhengrui Qin et al, *IMDGuard: Securing Implantable Medical Devices*

[10] C. C. Y. Poon, Y. T. Zhang, S. D. Bao (2006). *A Novel Biometrics Method to Secure Wire‐ less Body Area Sensor Networks for Telemedicine and M-Health,* IEEE Communication

[11] K. K. Venkatasubramanian, A. Banerjee, S. K. S. Gupta (2008). *Plenthysmogram-based secure inter-sensor communication in body sensor networks,* Proc. of IEEE Military Com‐

[12] K. K. Venkatasubramanian, A. Banerjee, S. K. S. Gupta (2010). *PSKA: usable and secure key agreement scheme for body area networks,* IEEE Transactions on Information Tech‐

[13] F. Miao, L. Jiang, Y. Li, Y. T. Zhang (2009). *Biometrics based novel key distribution solu‐ tion for body sensor networks,* Proc. Annual Conference of IEEE-EMBS, pp.2458−2461.

[14] A. Juels, M. Wattenberg (1999). *A fuzzy commitment scheme,* Proceedings of 6th ACM

[15] A. Juels, M. Sudan (2006). *A fuzzy vault scheme,* Design Codes and Cryptography, Vol.

[16] U. Uludag, S. Pankanti, A. K. Jain (2005). *Fuzzy vault for fingerprints,* In: Kanade T, Jai AK, Ratha NK. Proc. of the 5th Int'l Conf. on AVBPA. Berlin: Springer-Verlag, pp.

[17] Y. Wang, K. Plataniotis (2007). *Fuzzy vault for face based cryptographic key generation,* In:

[18] Y. Lee, K. Bae, S. Lee, K. Park, J. Kim (2007). *Biometric key binding: fuzzy vault based on iris images,* In: Lee SW, Li SZ eds. Proc. of the ICB 2007. LNCS 4642, Berlin: Springer-

Int'l. Conf. Eng. Med. and Bio. Soc., Shanghai, China.

information technology in biomedicine, Vol. 12, no. 6, pp. 772-779.

*with the External Wearable Guardian,* IEEE INFOCOM 2011.

nology in Biomedicine, Vol. 14, no. 1, pp.60-68.

conference on Computer and Communication Security.

Proc. of the Biometrics Symp. Berlin: Springer-Verlag, pp. 1−6.

pp.432−439.

Magazine, pp.73-81.

muniations, pp.1−7.

38, no. 2, pp. 237-257.

Verlag, pp.800−808.

310−319

For FDPS-based solution, the randomness performance of generated EIs, the computational complexity and the recognition rate pose great challenges to its application. Because the less satisfying randomness performance of EIs would bring about the security issue to the over‐ all solution, how to make generated EIs as random as possible while not affecting its recog‐ nition rate is an issue should be addressed. In addition, the high computational complexity especially the space complexity brought by large amount of chaff points should be de‐ creased to satisfy the stringent restriction of processing power, memory and energy for most sensor nodes. And what is the most important is that the recognition rate shall be signifi‐ cantly increased to make the solution applicable.

In some cases, not all of the physiological sensors that need to communicate with each other can obtain the needed information, such as IPI or same kind of physiological signals. Thus, how to extract a common feature from other kinds of physiological signal, such as respira‐ tion and blood pressure, might be further studied.

## **Author details**

Fen Miao, Shu-Di Bao and Ye Li

Key Laboratory for Biomedical Informatics and Health Engineering, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

## **References**


[5] D. Liu, P. Ning (2003). *Establishing pariwise keys in distributed sensor networks,* proceed‐ ings of the 10th ACM Conference on Computer and Communication, pp. 42-51.

crease the robustness performance of the solution. Secondly, a faster EI generation scheme based on minimum number of IPIs should be addressed to increase its real-time performance.

For FDPS-based solution, the randomness performance of generated EIs, the computational complexity and the recognition rate pose great challenges to its application. Because the less satisfying randomness performance of EIs would bring about the security issue to the over‐ all solution, how to make generated EIs as random as possible while not affecting its recog‐ nition rate is an issue should be addressed. In addition, the high computational complexity especially the space complexity brought by large amount of chaff points should be de‐ creased to satisfy the stringent restriction of processing power, memory and energy for most sensor nodes. And what is the most important is that the recognition rate shall be signifi‐

In some cases, not all of the physiological sensors that need to communicate with each other can obtain the needed information, such as IPI or same kind of physiological signals. Thus, how to extract a common feature from other kinds of physiological signal, such as respira‐

Key Laboratory for Biomedical Informatics and Health Engineering, Shenzhen Institutes of

[1] J. Hash, P. Bowen, A. Johnson, C. D. Smith, and D. I. Steinberg (2005). *An introductory resource guide for implementing the health insurance portability and accountability act (HI‐ PAA) security rule,* Nat. Inst Stand. Technol., NIST Spec. Publ. 800-66, Gaithersburg,

[2] The European Parliament and the council of The European Union (Jul. 2002). *Direc‐ tive 2002/58/EC concerning the processing of personal data and the protection of privacy in the electronic communications sector,* Official J. Eur. Communities, pp. L201/37-47.

[3] L. Eschenauer and V. Gligor (2002). *A key-management scheme for distributed sensor net‐ works,* Proceedings of the 9th ACM Conf. on Computer and Communication Security,

[4] H. Chan, A. Perrig, D. Song (2003). *Random key predistribution schemes for sensor net‐ works,* in: proceedings of the 2003 IEEE Symposium on security and privacy, May

Advanced Technology, Chinese Academy of Sciences, Shenzhen, China

cantly increased to make the solution applicable.

tion and blood pressure, might be further studied.

**Author details**

**References**

MD.

pp.41–47.

11-15, pp. 197-213.

Fen Miao, Shu-Di Bao and Ye Li

272 New Trends and Developments in Biometrics


[19] F. Miao, S. D. Bao, Y. Li (2010). *A Modified Fuzzy Vault Scheme for Biometrics-based Body Sensor Networks Security,* IEEE Globecom.

**Chapter 12**

**Influence of Skin Diseases on Fingerprint Quality and**

Fingerprint recognition belongs to one of the most often used biometric technologies world‐ wide. It is believed that fingerprints could be used for the recognition of a person in nearly any case; however there exist many cases, where the fingerprint recognition could not be used. There exist some influencing factors [1] that have an impact to the process of finger‐ print recognition, e.g. the environmental influences, dirtiness on finger or the sensor, elec‐ tromagnetic radiation or diseases. This chapter deals with the circumstances which influence the quality of fingerprints – we are limited on skin diseases here, further we ex‐

**•** *Fingerprint acquirement* – the fingerprint is scanned using a sensor (for sensor technologies see [1]), i.e. the physical human biometric attribute is digitized and transferred to the

**•** *Image enhancement* – this step is very important for further processing, because the quality of the fingerprint image could be enhanced here. There are several methods used for im‐ age quality enhancement – edge filters, filtering in frequency spectrum (after Fast Fourier

**•** *Thresholding* – the image is normally acquired with 256 gray levels, but we need a binary representation. Using various thresholding schemes (e.g. adaptive thresholding or region‐ al average thresholding), it is possible to separate papillary lines (ridges) from back‐

> © 2012 Dolezel et al.; licensee InTech. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

© 2012 Dolezel et al.; licensee InTech. This is a paper distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/3.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

**Recognition**

Michal Dolezel, Martin Drahansky,

http://dx.doi.org/10.5772/51992

**1. Introduction**

computer.

Transform), Gabor filter, etc.

ground (valleys).

Jaroslav Urbanek, Eva Brezinova and Tai-hoon Kim

plain how we can evaluate the quality of the acquired fingerprint.

The fingerprint recognition consists of five main steps (see Fig. 1) [2, 3, 4]:

Additional information is available at the end of the chapter

