An Approach to Speech and Iris based Multimodal Biometric System

Int. Journal of Electrical & Electronics Engg. Vol. 2, Spl. Issue 1 (2015) e-ISSN: 1694-2310 | p-ISSN: 1694-2426

NITTTR, Chandigarh EDIT -2015 176

An Approach to Speech and Iris basedMultimodal Biometric System

1SakshiSahore, 2TanviSood1M.Tech Student, 2Assistant Professor

1,2ECE Department, Chandigarh Engineering College, Ladran, [email protected], [email protected]

Abstract—Biometrics is the science and technology ofhuman identification and verification through the use offeature set extracted from the biological data of theindividual to be recognized. Unimodal and Multimodalsystems are the two modal systems which have beendeveloped so far. Unimodal biometric systems use asingle biometric trait but they face limitations in thesystem performance due to the presence of noise in data,interclass variations and spoof attacks. These problemscan be resolved by using multimodal biometrics whichrely on more than one biometric information to producebetter recognition results. This paper presents anoverview of the multimodal biometrics, various fusionlevels used in them and suggests the use of iris andspeech using score level fusion for a multimodalbiometric system.

Keywords—Biometric, unimodal, multimodal,recognition, score level fusion

I. INTRODUCTION

With the recent advancement in technology anddevelopment of electrically interconnected society, there isan essential requirement of accurate personalauthentication system to handle various personauthentication issues in daily life. There are severalauthentication systems that we use on daily basis such aspersonal identification number (PIN), smartcards andpasswords. These systems are possession based andknowledge based and can easily be misplaced, forgotten orforged [1]. To overcome these difficulties, biometricsystems for authentication are introduced. Biometrics is arobustious approach for the recognition of a person [2].Biometrics verify the identity of the subject based on afeature set extracted from the subject’s biologicalcharacteristics.Biometric characteristics are of two types:

Physiological: The characteristics related to the bodyof a person are called physiological characteristics.Fingerprints, face, iris, palm geometry, DNA are theexamples of the physiological characteristics. Thesecharacteristics do not change over time.

Behavioral: The characteristics related to the behaviorof a person are called behavioral characteristic. Voice,gait, signature and keystroke are the examples ofbehavioral characteristics. These are variant in nature.

A biometric system consists of two modes that areenrollment mode and authentication mode. In enrollmentmode, the biometric data of the subject is taken andprocessed for feature extraction. These features are usedfor the generation of template of that subject in which all

the feature variations are captured and stored in a database.During authentication mode, the features from the subjectto be identified are computed and then compared with thestored template in the database. If the features match, thesubject is recognized. Figure 1 shows a typical biometricsystem.

Biometric based person recognition system [3]

II. MODAL SYSTEM

A. Unimodal BiometricsA unimodal biometric system uses a single source ofbiometric information to generate the recognition result.Most of the deployed real world applications in biometricsare unimodal, that is, they use a single biometric trait forauthentication such as a biometric system based onfingerprints [4]. While unimodal biometric systems havesuccessfully been installed in various applications, butunimodal biometrics is still not fully solved problem [5].These systems a variety of issues like

Noisy data – The input biometric data might be noisyor the biometric sensors might be susceptible to noisewhich may lead to inaccurate matching and hencefalse rejection.

Intra-class variations – This occurs when the biometricdata acquired from an individual during verification isnot identical to the data stored in the template duringenrollment. This occurs due to incorrect interaction ofthe individual with the sensor.

Non-universality – Sometimes it is possible thatcertain individuals may not provide a particularbiometric causing failure to enroll (FTE).

Spoof attack –Unimodal biometrics are susceptible tospoof attacks where an imposter may attempt to fake








I. INTRODUCTION







II. MODAL SYSTEM













I. INTRODUCTION







II. MODAL SYSTEM







177 NITTTR, Chandigarh EDIT-2015

the biometric trait of an enrolled user in order tobypass the system.

The problems imposed in the unimodal biometrics limit theaccuracy and system performance.

B. Multimodal Biometrics

A multimodal biometric system uses more than onesource of biological data to generate the recognition result,for example a multimodal biometric system using iris andear [6]. Multimodal biometrics overcome the shortcomingsof unimodal biometrics [7] and is more reliable because ofthe use of more than one biometric trait and hence morepieces of information.

Multimodal scenarios: A multimodal biometric systemcan be designed to work in one of the following scnarios.

Multiple sensors: Highlight all author andaffiliation lines.The informationof the same biometric canbe acquiredby different sensors [8].The different samplesare thenprocessed by the same algorithm and the resultsarefused to get the resultant algorithm.

Multiple instances: The biometric information isextracted from the multiple instances of the same biometric[9].

Multiple algorithms: More than oneapproach/algorithm is used for feature extraction orclassification of the same biometric to improve the systemperformance [10].

Multiple biometric:Evidence from the multiplebiometric characteristics is taken [11].

Multiple samples:Multiple samples are acquiredfrom the same biometric by a single sensor and processedby the same algorithm to obtain the recognition results[12].

Multimodal Biometric System [13]

Fusion Levels in Multimodal Biometrics: Whiledesigning a multimodal system, different fusion strategiescan be used to integrate the biometric data.

Fusion Levels in Multimodal Biometrics

Sensor level fusion:Highlight all author andaffiliation lines. Sensor level fusion is done in a systemsystem using multiple sensors or in a system using a singlesensor at multiple instances. In this, the biometric dataobtained by the sensor is combined.

Feature level fusion:Feature level fusion is doneby extracting the features of different biometric sourcesindividually and then combining those features into asingle feature vector [14].

Score level fusion:Score level fusion is performedby individually processing (sensing and extractingfeatures) different biometric sources and finding theirmatch scores. These scores are then combined to makeclassification [15].

Decision level fusion:After each biometric sourceis processed and recognition decision is made for eachbiometric data, fusion is executed at the decision level[16].

III. SCORE LEVEL FUSION

Score level fusion is the most popular and commonapproach in the multimodal biometrics system due itssimple procedure. Matching scores contain richinformation about the input pattern. Each classifiersprovides a matching scores and scores of differentclassifiers are combined to produce the final score.

A. Fusion algorithms

When different matching scores of different biometrics areacquired, their fusion is done. For the fusion of thematching scores, different algorithms can be applied.These algorithms include product rule, sum rule, max ruleand min rule.

Consider ( ⃗) as the output of individual classifiers,as a feature vector to ith classifier, R as the number of

different classifiers and be the output. The different rulescan be applied as

BIOMETRICFUSION

BEFOREMATCHIG

AFTERMATCHING

SCORELEVEL

FEATURELEVEL

SCORELEVEL

DECISION LEVEL


























BIOMETRICFUSION

BEFOREMATCHIG

AFTERMATCHING

SCORELEVEL

FEATURELEVEL

SCORELEVEL

DECISION LEVEL


























BIOMETRICFUSION

BEFOREMATCHIG

AFTERMATCHING

SCORELEVEL

FEATURELEVEL

SCORELEVEL

DECISION LEVEL



Score Level Fusion in Multimodal Biometric System

Product rule:Different biometric traits of an individual(such as iris, fingerprints and ear) are mutuallyindependent and the product rule is applied based on this.

= ( ⃗)Sum rule: The sum rule takes the scores of the

individual classifiers to simply calculate their sum.

= ( ⃗)Max rule:The max rule approximates the output by the

maximum value of the scores.= max ( ⃗)Min rule:The max rule approximates the output by the

minimum value of the scores.= min ( ⃗)B. NormalizationNormalization is done after determining the matchingscores from different biometrics. Score normalization isessential because the matching scores of differentbiometrics are obtained from different algorithms andhence may not have the same underlying properties, that is,they may be of different nature and scale. Normalizationchanges the scale of the different scores and brings them toa common domain. After normalization, the scores arecombined. The most common normalization algorithmsused are

If S = ( , , ,… ,……. ) is a vector of M scores, thenthe normalization score, will be

Min max normalization:

=µ( )( ) ( )

Where max(s) = maximum value of raw score andmin(s) = minimum value of raw score

Z-score normalization

=µ( )( )

Where µ(s) = mean deviation of set of score vectorsand σ(s) = standard deviation of the set of score vectors

Median-MAD normalization

=

Where MAD = median (| − |)Tanh normalization

= 0.5 ℎ 0.01 − µ( )( ) + 1Where µ( )= mean deviation calculated from scores

and ( ) = standard deviation calculated from scores.

IV. MULIIMODAL BIOMETRIC USING IRIS ANDSPEECH

As mentioned earlier, there are two types of biometriccharacteristics in human beings, physiologicalcharacteristics and behavioral characteristics. With theselection of appropriate modals and fusion scheme,optimal results can be achieved. There are severalinspirations to choose iris and speech for a multimodalbiometric system. Iris is a physiological trait while speechis a behavioral trait. These two biometrics can becombined to form an effective multimodal biometricsystem. Iris recognition requires small high qualitycameras for operating and processes the output in 1 to 2seconds. Iris patterns carry astonishing amount ofinformation and remain unchanged throughout theindividual’s lifetime. Iris recognition suffers no problemwith eyeglasses and contact lenses. It is hence, one of themost stable and precise personal identification biometricwhich gives excellent recognition performance [13] [17].Voice recognition system is an emerging biometrictechnology. Voice is usually considered as a behavioralcharacteristic but it is actually a combination of bothphysiological and behavioral characteristics. Thephysiological part of the voice remains invariant while thebehavioral part changes over time depending on the age,medication and emotional state of an individual [18].Voice recognition biometric system is typically cheap withthe requirement of a microphone. It has high userpreference and the processing speed of 5 seconds with highefficiency [18].

MATCHSCORE

MATCHING

MATCHSCORE

FEATUREVECTOR

MATCHING

FUSION

TOTALSCORE

DECISION

TEMPLATE

TEMPLATE

SENSORDATA

SENSOR

DATA

FEATUREEXTRACTI

ON

FEATUREEXTRACTI

ON

FEATUREVECTOR



V. CONCLUSION AND FUTURE SCOPE

The performance of the unimodal biometric recognitionsystems suffer from several limitations that can beovercome by the use of multimodal biometrics.Multimodal biometrics combine the information obtainedfrom the different sources through the use of an effectivefusion scheme. Multimodal biometric systems work indifferent scenarios and different fusion levels. Theperformance of a multimodal biometric system can beimproved through the selection of appropriate fusionscheme. In this paper, the modality of iris and speecharesuggested with their score level fusion due to its simpleprocedure and rich information.

REFERENCES[1] Arun Ross and Anil Jain, “Information Fusion in Biometrics”,

Pattern Recognition letters, vol. 2, issue 13, pp. 2115-2125, 2003[2] Suma Swany and K. V. Ramakrishnan, “An efficient speech

recognition system”, Computer Science and Engeneering : Aninternational journal (CSEIT), vol. 3, no. 4, Aug 2013

[3] R. Frischholz, U. Dieckmann,“BiolD: A multimodal biometricidentification system”, Computer, Vol. 33,No. 2, pp. 64-68,2000

[4] Sravya V., Radha Krishna Murthy, RavindraBabuKallam, SrujanaB., “A survey on fingerprint biometric system”, InternationalJournal of Advanced Research in Computer Science and SoftwareEngineering, vol. 2, issue 4, April 2012

[5] SahilPrabhakar, SharathPankanti, Anil K. Jain, “Biometricrecognition: Security and privacy concerns”, Security and Privacy,IEEE, vol. 1, issue 2, pp. 33-42, 2003

[6] M. Fatima Naddheen, S. Poornima, “Fusion in MultimodalBiometric using Iris and Ear”, Proceedings of IEEE Conference onInformation and Communication Technologies, pp. 83-87, 2013

[7] KomalSondhi, YogeshBansal, “Concept of Unimodal andMultimodal Biometric systems”, International Journal of AdvancedResearch in Computer Science and Software Engineering, vol. 4,issue 6, 2014

[8] ThirimachosBourlai, Nathan Kalka, Arun Ross, BojanCukic,Lawrence Hornak, “ Cross Spectral Face Verification in the ShortWave Infrared (SWIR) Band”, Proc. of International Conference onPattern Recognition, IEEE, pp. 1343-1347, 2010

[9] DzatiAthiarRamli, Nurue Hayat Che Rani, KhairulAnuarIshak,“Performances of Weighted Sum Rule Fusion scheme in Multi-instance and Multimodal Biometric system”, World AppliedScience Journal, vol. 12, no. 11, pp. 2160-2167, 2011

[10] Vaidehi V., Teena Mary Tressa, NareshBabu N. T., AnnisFathimaA., Balamurali P., Girish Chandra M., “Multi Algorithmic FaceAuthentication System”, Proceedings of the International Multi-conference of Engineers and Computer Scientists, vol. 1, pp.485-490, 2011

[11] GandhimathiAmirthalingam, Radhamani G., “A MultimodalApproach for Face and Ear Biometric System”, International Journalof Computer Science Issues (IJCSI), vol.10, issue 5, no. 2, pp. 234-240, 2013

[12] Xi Cheng, Sergey Tulvakov, VenGovindaraju, “Combination ofMultiple Samples Utilizing Identification Modal in BiometricSystem”, International Conference on Biometric Compendium,IEEE, pp. 1-5, 2011

[13] Anil K. Jain, Arun Ross, SahilPrabhakar, “An Introduction toBiometric Recognition”, IEEE Transactions on Circuits andSystems for Video Technology, vol. 14, no. 1, pp. 4-20, 2004

[14] Vincenzo Conti, Carmelo Militello, FlippoSorbello, “A FrequencyBased Approach for Feature Fusionin Fingerprint and Iris

Multimodal Biometric Identification Systems”, IEEE Transactionson Systems, Man and Cybernetics, vol. 40, no. 4, pp. 384-395, 2010

[15] Sayed Hassan Sadeghzadeh, MortezaAmirsheibani,AnsehDaneshArasteh, “Fingerprint and Speech Fusion: AMultimodal Biometric System”, International Journal of ElectronicsCommunication and Computer Technology (IJECCT), vol. 4, no. 2,pp. 570-576, 2011

[16] Kihal N., Chitroub S., Meunier J., “Fusion of Iris and Palmprint forMultimodal Biometric Authentication,” 4th International Conferenceon Image Processing theory, tools and applications (IPTA), 2014

[17] A.K. Jain, A. Ross, S. Pankanti, “ Biometrics, a tool forinformation”, IEEE Transactions on Information Forensics andsecurity, vol. 1, issue 2, pp. 125-143, 2006

[18] DwijenRudrapal, Smita Das, S. Debbarama, N. Kar, N. Debbarama,“Voice recognition and authenticationas a proficient biometric tooland its application in online exam for PH people”, InternationalJouranal of Computer Applications, vol. 39, no. 12, pp. 6-12, 2012

[19] Suma Swamy and K. V. Ramakrishnan, “An Efficient SpeechRecognition System”, Computer Science and Engineering: AnInternational Journal (CSEIJ), vol. 3, no. 4, pp. 21-27, Aug 2013.

An Approach to Speech and Iris based Multimodal Biometric System

Engineering

biometric characteristics

multimodal biometric

biometric data

unimodal biometric systems

typical biometric system

biometric information

multimodal systems

authentication systems