Fuzzy Logic and Multi-biometric Fusion

An Overview

Fabian Maul and Naser Damer

Identiﬁcation and Biometrics, Fraunhofer IGD, Fraunhoferstraße 5, Darmstadt, Germany

Keywords:

Multi-Biometric Fusion, Fuzzy Logic, Fuzzy Fusion.

Abstract:

Fuzzy logic has been proposed to improve various aspects of multi-biometric applications including enhance-

ments to the decision making of the application and the robustness to noisy data. This paper discusses recent

work that utilized fuzzy logic techniques within the multi-biometric fusion problem. This discussion is pre-

sented under two categories, the type of authentication scenario and the nature of the fused data. The paper

also presents an introduction to fuzzy logic and multi-biometric fusion. Based on the discussed works, this

paper aims to establish current trends and research possibilities in this ﬁeld.

1 INTRODUCTION

In order to improve biometric recognition, different

types of multi-biometrics haven been studied. The

multi-biometrics can be categorized into following

types: multi-algorithm, multi-instance, multi-modal,

multi-sample and multi-sensor biometrics. The cate-

gory most relevant for this paper is multi-modal bio-

metrics. In multi-modal biometrics, two or more bio-

metric characteristics are used. E.g. the work of Lau

et al. (Lau et al., 2004) and Hui et al. (Hui et al.,

2007) made use of three characteristics: face, ﬁn-

gerprint and speech. The process of combining the

results of multi-biometrics is called fusion. Ross et

al. (Ross et al., 2008) name advantages of multi-

biometrics in the areas of: universality, indexing,

robustness, resistance to noisy data, continuity and

fault tolerance. Based on these advantages other re-

search has shown that multi-biometrics can improve

performance of biometric recognition (Chang et al.,

2003) (Hong et al., 1999).

Fuzzy logic is a way of reasoning by approxima-

tion that deals with vague or uncertain data by assign-

ing them truth values between 0 and 1 (Zadeh, 1965).

This allows the modelling of vague data, where for

example a temperature could be 0.3 warm and 0.7

cold corresponding to the vague expression of “fairly

cold“. This is done in three steps. First each in-

put is assigned to a linguistic variable. This step is

called fuzziﬁcation. Then for each variable member-

ship functions with different degrees of membership

are obtained. Next, fuzzy rules can be utilized where

logic can be applied to these membership functions.

E.g. in the form of “If condition-1 and condition-2

then decision a“. This decision often is used as the

output for the fuzzy system. The process of convert-

ing this fuzzy output to a continuous value is called

defuzziﬁcation.

Fuzzy logic can be used in biometrics to deal with

the quality of the samples of characteristics, which

can be affected by lighting, noise and user-device in-

teractions. It can also be used to perform the fusion

and therefore enhance the decision making of a multi-

biometric application. Another idea is to use this

principle for continuous authentication. Continuous

authentication aims to have the individual authenti-

cated using an ongoing identiﬁcation or veriﬁcation

process. E.g. while the subject is using a PC or while

it is inside a speciﬁc area of a building.

In the next two sections further details along with

the advantages of fuzzy logic and multi-modal fu-

sion will be presented. The section after will present

the research done in order to combine these two dis-

ciplines. Finally, a conclusion will be given where

trends and research opportunities will be discussed.

2 FUZZY LOGIC

An application using fuzzy logic has the advantage

of being able to deal with both numerical as well as

linguistic data. This can be of advantage when deal-

ing with vague or incomplete data. Being able to deal

with data referring to concepts such as “high conﬁ-

218

Maul F. and Damer N..

Fuzzy Logic and Multi-biometric Fusion - An Overview.

DOI: 10.5220/0005193302180222

In Proceedings of the International Conference on Pattern Recognition Applications and Methods (ICPRAM-2015), pages 218-222

ISBN: 978-989-758-076-5

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

dence“, “low conﬁdence“ or “quality“ offers possibil-

ities in both decision making as well as coping with

external factors of an application.

Mendel describes a fuzzy logic system (FLS), see

Figure 1, as follows (Mendel, 1995): An FLS can be

viewed as a mapping from inputs to outputs. Rules

may be provided by experts or can be extracted from

numerical data. In either case, these rules are ex-

pressed as a collection of “If Then“ statements, e.g.

“IF u

is very warm and u

quite low, THEN turn v

somewhat to the right.“ The fuzziﬁer maps the crisp

input to fuzzy sets. This is needed to work with the

aforementioned rules. The inference engine of the

FLS maps fuzzy sets to fuzzy sets. It handles the way

in which rules are combined, which helps in making

decisions. Finally, the defuzziﬁer maps the output sets

into crisp numbers.

George et al. state that fuzzy logic is used in the ar-

eas of civil engineering, mechanical engineering, in-

dustrial engineering, computer engineering, robotics

and in reliability theory (George J and Bo, 2008). Ex-

amples for such an approach can be found in a review

by Chen et al., where different ways of using fuzzy

logic to improve the ability of dealing with vague or

unclear data on the semantic web are shown (Chen

et al., 2012). Furthermore Mendel names a list of

real world applications, that make use of fuzzy logic

(Mendel, 1995). These applications are used for con-

trol (e.g. space shuttle docking (NASA)), schedul-

ing and optimization (e.g. stock market analysis (Ya-

maichi Securities)) and signal analysis for tuning and

interpretation (e.g. TV picture adjustment (Sony)).

3 MULTI-MODAL FUSION

A multi-modal biometric system is a multi-biometric

system where more than one biometric characteristic

is used to recognize a subject. In order to make a

ﬁnal decision regarding this recognition, the different

information need to be combined or merged (fusion).

There are different ways to implement such a fusion

process. Ross et al. name the following (Ross et al.,

2008):

Decision-level Fusion. In decision-level fusion the

decisions of the uni-biometric systems are used to

compute the ﬁnal decision. E.g. using a “AND“

and “OR“ rule-set.

Rank-level Fusion. If the biometric system is used

for identiﬁcation, the output can be viewed as a

ranking of the enrolled identities. The goal of

rank-level fusion is to consolidate this output by

the individual biometric subsystems in order to

derive a consensus rank for each identity. Which

then can be used to build the ﬁnal decision.

Score-level Fusion. In score-level fusion the com-

parison scores output of the different biomet-

ric comparators are combined to generate a new

match score that can be subsequently used by the

veriﬁcation or identiﬁcation modules for render-

ing the ﬁnal decision.

Feature-level Fusion. Using this fusion method, the

feature sets originating from multiple different

biometric algorithms are consolidated into a sin-

gle feature set. This is achieved by the application

of appropriate feature normalization, transforma-

tion and reduction schemes.

Sensor-level Fusion. This fusion method refers to

the consolidation of either raw data obtained us-

ing multiple sensors, or multiple snapshots of a

biometric characteristic using a single sensor.

Approaches using fuzzy logic to improve multi-

modal fusion focused mainly on decision-level fusion

and score-level fusion.

4 LITERATURE SURVEY

4.1 Classiﬁcation Criteria

In this section, previous research work using fuzzy

logic to improve multi-biometric fusion are presented

under two discussion categories. The ﬁrst category

differentiates the proposed systems based on the au-

thentication scenario intended, weather it is a conven-

tional (static) or continuous authentication scenario.

The second part will present the previous work de-

pending on the kind of information that was fused and

used fuzzy logic techniques to achieve their goals.

4.2 Type of Authentication

4.2.1 Static Authentication

Conventional or static authentication is the identiﬁ-

cation or veriﬁcation scenario common for biometric

systems where the recognition decision is built only

once and subsequent actions are performed based on

that decision.

In the approach proposed by Park et al. (Park et al.,

2006), fuzzy logic is utilized in the decision making

level of the system in order to improve the veriﬁcation

rate and to lower the FAR and FRR. The proposed

system fused information gathered from a face com-

parator and a speaker comparator. Unsupervised prin-

cipal component analysis (PCA) was used for face

FuzzyLogicandMulti-biometricFusion-AnOverview

219

Figure 1: overview of a typical fuzzy logic system (Mendel, 1995).

Table 1: Table of work that was used for static authentication. Ordered by mention in this paper.

Multi-modal human veriﬁcation using face and speech (Park et al., 2006)

Fuzzy logic decision fusion in a multimodal biometric system (Lau et al., 2004)

Adaptive weight estimation in multi-biometric veriﬁcation using fuzzy logic de-

cision fusion

(Hui et al., 2007)

Quality based Speaker Veriﬁcation Systems using Fuzzy Inference Fusion

Scheme

(Hamid and Ramli, 2014)

An efﬁcient multi-modal biometric person authentication system using Fuzzy

Logic

(Vasuhi et al., 2010)

Multimodal biometric system fusion using ﬁngerprint and iris with fuzzy logic (Abdolahi et al., 2013)

recognition (Rowley et al., 1998). For speaker ver-

iﬁcation Hidden Markov Models (HMM) (Samaria

and Young, 1994) were used by the authors. The ex-

periments carried out by the authors achieved a veriﬁ-

cation rate of 99.99% at an FAR of 0.001%, which is

an improvement over the individual modalities’ per-

formance of 98.5% and 97.37% veriﬁcation rate for

face and speaker comparators respectively, both at an

FAR of 0.01%.

Lau et al. also proposed a system where fuzzy

logic is used to improve a veriﬁcation system (Lau

et al., 2004). The authors combined ﬁngerprint, face

and speaker veriﬁcation into a system using fuzzy

logic to dynamically alter the weights of the differ-

ent characteristics. External factors, e.g. lighting, are

measured and then used by the fuzzy logic system to

determine the weights for the ﬁnal decision calcula-

tion. Only the weights for face and ﬁngerprint recog-

nition were calculated using the fuzzy logic based

approach. In different experiments the proposed ap-

proach achieved EER values in the range of 0.31% to

0.81% compared to the base line fusion by weighted

avarage scores with EER values in the range of 0.50%

and 0.84%.

Hui et al. presented an improvement over the sys-

tem proposed by Lau et. al. by adding a method to

compute a measurement of noise in the speaker recog-

nition (Hui et al., 2007). This signal-to-noise ratio is

then used by the fuzzy logic system to calculate the

optimized weight for the speaker comparison. In ex-

periments utilizing paired test sets the proposed so-

lution was able to show an overall improvement of

the EER by 42.1% compared to a solution based on

weighted average fusion.

A similar system, where fuzzy logic is used to

compute weights for the different biometric scores,

was proposed by Hamid et al. (Hamid and Ramli,

2014). The authors performed experiments using both

the Sugeno-type and Mamdani-type fuzzy models.

The paper did not present a comparison to base line

fusion solutions, however, it clearly showed the per-

formance gained by the fuzzy fusion approach com-

pared to single modalities.

Vasuhi et al. proposed a system where the ﬁnal

decision is made by fusing the comparison scores

produced by speaker and ﬁngerprint matchers with

fuzzy logic rules (Vasuhi et al., 2010). For voice

recognition, the authors used a text dependent speaker

veriﬁcation using HMM. For the feature extrac-

tion and classiﬁcation Mel-frequency Cepstral Coef-

ﬁcients (MFFCC) were used (Hasan et al., 2004)

(Biswas et al., 2007). The authors proposed and used

the Cross Correlation of Field Orientation (CCFP) as

bases for their ﬁngerprint comparator. The work did

not present any statistically signiﬁcant results, how-

ever it claims to present an efﬁcient solution that over-

comes the drawbacks of individual sensors.

Abdolahi et al. follow a similar approach, where

ICPRAM2015-InternationalConferenceonPatternRecognitionApplicationsandMethods

220

Table 2: Table of work that was used for continuous authentication. Ordered by mention in this paper.

Ensuring the identity of a user in time: a multi-modal fuzzy approach (Azzini et al., 2007)

A fuzzy approach to multimodal biometric continuous authentication (Azzini et al., 2008)

Continuous authentication using mouse dynamics (Mondal and Bours, 2013)

the scores of ﬁngerprint and iris matchers are used as

inputs for the fuzzy logic system, which calculates the

ﬁnal decision (Abdolahi et al., 2013). Discussed re-

sults only show advantages with respect to uni-modal

base comparators.

4.2.2 Continuous Authentication

In highly sensitive environments, it might be neces-

sary to conﬁrm the identity of subjects at random in-

tervals to prevent identity substitution after the initial

authentication. For these cases, systems using contin-

uous authentication were proposed.

Two papers by Azzini et al. focus on the integra-

tion of fuzzy logic into a solution for continuous au-

thentication (Azzini et al., 2007) (Azzini et al., 2008).

In both systems, fuzzy logic was used to calculate the

decisions using the scores scored by different biomet-

ric sources as its input. The authors conclude their

work with a positive view on the achieved perfor-

mance with emphasis on the capabilities of the solu-

tion when the context variables in the acquisition are

not optimal.

In theory, fuzzy logic techniques can be used to

build a more robust trust model for continuous au-

thentication. Recent advances regarding continuous

authentication and trust models can be seen in the

work of Mondal et al. (Mondal and Bours, 2013).

4.3 Type of Data Fused

4.3.1 Fuzzy Logic Decision

Here, the input for the fuzzy logic system is the bio-

metric scores from different sources. The fuzzy logic

solution tries to optimally fuse those scores into a ﬁ-

nal uniﬁed biometric decision. Much previous work

followed this approach to improve the performance

of biometric systems. The papers by Abdolahi et

al., Azzini et al., Park et al. and Vasuhi et al. (Ab-

dolahi et al., 2013) (Azzini et al., 2007) (Azzini et al.,

2008) (Park et al., 2006) (Vasuhi et al., 2010) were

based on this technique. More details about those

works and the achieved performances are discussed

in Section 4.2.

4.3.2 Fuzzy Adaptive Weight Score-level Fusion

For this approach, fuzzy logic is used to calculate

optimal weights for the scores of different biometric

sources using external factors such as capture quality

measures (e.g. illumination). The fusion itself is a

weighted score-level fusion. This methodology was

used by Hamid et al., Hui et al. and Lau et al. (Hamid

and Ramli, 2014) (Hui et al., 2007) (Lau et al., 2004).

4.3.3 Fusion Using Non-biometric Data

Azzani et al. included a strictly non-biometric input

into the fuzzy logic fusion (Azzini et al., 2007) (Azz-

ini et al., 2008). A password entered by the subject

is fused along with the biometric information to pro-

duce a ﬁnal authentication decision. While the score

of the password is simply represented by a 1 if it was

correct, and a 0, if it was not correct, this inclusion

was noteworthy and can be extended to different non-

biometric information.

5 CONCLUSIONS

This paper presented a literature survey of the utiliza-

tion of fuzzy logic within multi-biometric systems. It

discusses the previously presented work based on two

criteria. First is the authentication scenario, whether

it is a conventional (static) or a continuous authenti-

cation. The second criterion is based on the type of

fused information as well as the nature of the fuzzy

logic application. All reviewed papers show that the

use of fuzzy logic can improve multi-biometric sys-

tems in certain scenarios. The presented works, es-

pecially the ones conducted in (Azzini et al., 2008),

show that selecting appropriate fuzzy membership

functions and deﬁning the most beneﬁcial set of fuzzy

rules is not a trivial task and might require time and

effort. These parameters also depend strongly on the

application in question. Unfortunately none of the ex-

periments in the previous work were conducted us-

ing a large and publicly available database. Future

work should, therefore, be more comparable to exist-

ing solutions and include the investigation of the ap-

propriate scenarios for the utilization of fuzzy logic.

It should as well discuss methods that should be fol-

lowed to optimize the parameters of the fuzzy logic

FuzzyLogicandMulti-biometricFusion-AnOverview

221

solution. Another task for future work would be the

veriﬁcation of these approaches by using large and

open accessibly database. Reproducibility and veri-

ﬁcation of the results these proposals have produced

might support a trend of continuing and expanding re-

search in combining fuzzy logic with biometric recog-

nition.

REFERENCES

Abdolahi, M., Mohamadi, M., and Jafari, M. (2013). Mul-

timodal biometric system fusion using ﬁngerprint and

iris with fuzzy logic. International Journal of Soft

Computing and Engineering, 2(6):504–510.

Azzini, A., Damiani, E., and Marrara, S. (2007). Ensur-

ing the identity of a user in time: a multi-modal fuzzy

approach. In IEEE International Conference on Com-

putational Intelligence for Measurement Systems and

Applications (CIMSA), pages 94–99. IEEE.

Azzini, A., Marrara, S., Sassi, R., and Scotti, F. (2008).

A fuzzy approach to multimodal biometric continu-

ous authentication. Fuzzy Optimization and Decision

Making, 7(3):243–256.

Biswas, S., Ahmad, S., and Molla, M. K. (2007). Speaker

identiﬁcation using cepstral based features and dis-

crete hidden markov model. In Information and

Communication Technology, 2007. ICICT’07. Inter-

national Conference on, pages 303–306. IEEE.

Chang, K., Bowyer, K. W., Sarkar, S., and Victor, B.

(2003). Comparison and combination of ear and face

images in appearance-based biometrics. Pattern Anal-

ysis and Machine Intelligence, IEEE Transactions on,

25(9):1160–1165.

Chen, H., Wu, Z., and Cudr

e-Mauroux, P. (2012). Seman-

tic web meets computational intelligence: State of the

art and perspectives [review article]. Computational

Intelligence Magazine, IEEE, 7(2):67–74.

George J, K. and Bo, Y. (2008). Fuzzy sets and fuzzy logic,

theory and applications. -.

Hamid, L. and Ramli, D. (2014). Quality based speaker ver-

iﬁcation systems using fuzzy inference fusion scheme.

In International Conference on Communications, Sig-

nal Processing and Computers.

Hasan, M. R., Jamil, M., and Rahman, M. G. R. M. S.

(2004). Speaker identiﬁcation using mel frequency

cepstral coefﬁcients. variations, 1:4.

Hong, L., Jain, A. K., and Pankanti, S. (1999). Can multi-

biometrics improve performance? In Proceedings Au-

toID, volume 99, pages 59–64.

Hui, H. P.-S., Meng, H. M., and Mak, M.-W. (2007).

Adaptive weight estimation in multi-biometric veriﬁ-

cation using fuzzy logic decision fusion. In Acoustics,

Speech and Signal Processing, 2007. ICASSP 2007.

IEEE International Conference on, volume 1, pages

I–501. IEEE.

Lau, C. W., Ma, B., Meng, H. M.-L., Moon, Y.-S., and Yam,

Y. (2004). Fuzzy logic decision fusion in a multimodal

biometric system. In INTERSPEECH.

Mendel, J. M. (1995). Fuzzy logic systems for engineering:

a tutorial. Proceedings of the IEEE, 83(3):345–377.

Mondal, S. and Bours, P. (2013). Continuous authentication

using mouse dynamics. In Biometrics Special Inter-

est Group (BIOSIG), 2013 International Conference

of the, pages 1–12. IEEE.

Park, C., Paik, J., Choi, T., Kim, S., Kim, Y., and Namkung,

J. (2006). Multi-modal human veriﬁcation using

face and speech. In Computer Vision Systems, 2006

ICVS’06. IEEE International Conference on, pages

54–54. IEEE.

Ross, A., Nandakumar, K., and Jain, A. K. (2008). Intro-

duction to multibiometrics. In Handbook of biomet-

rics, pages 271–292. Springer.

Rowley, H. A., Baluja, S., and Kanade, T. (1998). Neu-

ral network-based face detection. Pattern Analy-

sis and Machine Intelligence, IEEE Transactions on,

20(1):23–38.

Samaria, F. and Young, S. (1994). HMM-based architecture

for face identiﬁcation. Image and vision computing,

12(8):537–543.

Vasuhi, S., Vaidehi, V., Babu, N. N., and Treesa, T. M.

(2010). An efﬁcient multi-modal biometric person au-

thentication system using fuzzy logic. In Advanced

Computing (ICoAC), 2010 Second International Con-

ference on, pages 74–81. IEEE.

Zadeh, L. A. (1965). Fuzzy sets. Information and control,

8(3):338–353.

ICPRAM2015-InternationalConferenceonPatternRecognitionApplicationsandMethods

222