International Journal of Engineering and Technical Research (IJETR)

ISSN: 2321-0869, Volume-2, Issue-8, August 2014

A Multibiometric Recognition System :

Categorization, Retrieval Techniques & Applications
Kshama Dwivedi, Sachin Singh Thakur, Mayur Shishupal, Aarti Bhirud
From an application point of view, following properties
Abstract Reliable human authentication schemes are of should also be taken into account.
paramount importance in our highly networked society. The
multi-biometric recognition system enhances the performance Performance: The required recognition accuracy in an
and accuracy by consolidating the benefits of various uni-modal application should be achievable using the
biometric systems. They are expected to meet the stringent
characteristics.
performance requirements imposed by large-scale
authentication systems. This paper presents the various
Acceptability: Acceptability refers to the willingness
categories of multimodal system, the techniques index codes, by the subject to present his biometric
Gittins index algorithm, used for pattern retrieval from large characteristics.
database & its applications. Spoof Resistance: This refers to how difficult it is to
use artifacts (for example, fake fingers) in case of
Index Terms Multibiometric, unimodal, biometric, index physiological characteristics and mimicry in case of
codes, index algorithm behavioral characteristics.

I. INTRODUCTION A number of biometric characteristics exist and are in use

Biometrics refers to metrics related to human characteristics in various applications. Each biometric has its strengths and
and traits. Biometric identification (or biometric weaknesses, and the choice depends on the application. No
authentication) is used in computer science as a form of single biometric is expected to effectively meet the
identification and access control. It is also used to identify requirements of all the applications. In other words, no
individuals in groups that are under surveillance. Biometric biometric is optimal. The match between a specific
identifiers are often categorized as physiological versus biometric and an application is determined depending upon
behavioral characteristics. the operational mode of the application and the properties of
the biometric characteristic. A brief introduction to the
Physiological characteristics are related to the shape of the commonly used biometrics is given below.
body. Examples include, but are not limited to fingerprint,
palm print, face recognition, DNA, hand geometry, iris
recognition, retina and odor/scent. Behavioral characteristics
are related to the pattern of behavior of a person, including but
not limited to typing rhythm, gait, and voice.
Some desirable properties of biometric characteristics for
good discrimination and reliable recognition performance are
as follows:
Universality: Every individual should possess the
Fig. (a)
characteristic.
Uniqueness: The characteristics should be sufficiently Fingerprints:
distinguishable across individuals comprising the Fingerprints, as shown in Fig. (a) are unique and consistent
population. over time and hence being used since a long time. A
Permanence: The biometric characteristics should be fingerprint is a pattern of ridges and valleys on the surface of a
sufficiently invariant over a period of time. fingertip. Ridges are the upper skin layer segments of the
Measurability: It should be possible to acquire the finger and valleys are the lower segments. The various kinds
characteristics without causing undue of discontinuities in ridges (minutiae) have sufficient
inconvenience. The acquired raw data should be discriminatory information to recognize fingerprints. Ridge
suitable for further processing. bifurcation (where the ridge splits) and ridge ending (where
the ridge ends) are the important minutiae points.
Minutiae-based fingerprint recognition usually represents
fingerprint by these two ridge characteristics called as
Manuscript received August 19, 2014. minutiae. The uniqueness of a fingerprint can be determined
Kshama Dwivedi, pursing ME (Computer Engineering) ARMIET, by the pattern of ridges and furrows as well as the minutiae
Asangaon (Mumbai), India
Sachin Singh Thakur, pursuing MTech(Computer Science) T.I.T, points. Availability of multiple fingerprints of a person makes
Bhopal (M.P.), India fingerprint recognition suitable for use in large-scale
Mayur Shishupal, - pursuing ME (Electronics & Telecommunication) identification involving millions of identities. However, the
Y.T.I.E.T. Bhivpuri (Mumbai), India problem with the large scale fingerprint recognition system is
Aarti Bhirud, - pursuing ME (Computer Engineering) Y.T.I.E.T.
Bhivpuri (Mumbai), India.

78 www.erpublication.org
 A Multibiometric Recognition System : Categorization, Retrieval Techniques & Applications

the requirement of huge amount of computational resources, predetermined part of the retinal vasculature could be imaged.
especially in the identification mode. The image acquisition involves cooperation of the subject,
entails contact with the eyepiece, and requires a conscious
Face: effort on the part of the user. All these factors adversely affect
Face recognition as shown in Fig (b) is a non-intrusive method the public acceptability of retinal biometric. Retinal
and also requires minimum cooperation from the subject. The vasculature can reveal some medical conditions, e.g.,
dimensions, proportions and physical attributes of a persons hypertension, which is another factor deterring the public
face are unique. In some application scenario like crowd acceptance of retinalscan-based biometrics.
surveillance, face recognition probably is the only feasible
modality to be used. Face recognition can be in a static
controlled environment or a dynamic uncontrolled
environment. One popular approach to face recognition is
based on the location, dimensions and proportions of facial
attributes such as eyes, eyebrows, nose, lips, and chin and
their spatial relationships. Another approach being widely
used is based on the overall analysis of the face image that
represents face as a weighted combination of a number of
canonical faces.
Face recognition involves two major tasks: i) face location
and ii) face recognition. Face location is determining the
location of face in the input image. For recognizing the Fig (c)
located face, the eigenface approach is one of the very popular Ear:
methods. The eigenface-based recognition method consists of It has been suggested that the shape of the ear and the
two stages: i) training stage and ii) operational stage. In the structure of the cartilaginous tissue of the pinna are
training stage, training set of face images are acquired. The distinctive. The ear recognition as shown in the Fig (d)
acquired face images are projected into lower dimensional approaches are based on matching the distance of salient
subspace using Principle Component Analysis (PCA). A set points on the pinna from a landmark location on the ear. The
of images that best describe the distribution of training images features of an ear are not expected to be very distinctive in
in a lower dimensional face space (the eigenspace) is establishing the identity of an individual.
computed. Then the training facial images are projected into
this eigenspace to generate representation of the training
images in the eigenspace. In the operational stage, the input
face image is projected into the same eigenspace that the
training samples were projected into. Then, recognition can
be performed by a classifier operating in the eigenspace.

Fig (b)
Fig (d)
Iris and Retina: DNA:
Fig (c) show the difference between iris and retina scan. Iris is As shown in the Fig (e) Deoxyribonucleic acid (DNA) is the
the annular region of the eye regulating the size of the pupil. It one-dimensional (1D) ultimate unique code for ones
is bounded by pupil and sclera (white of the eye) on either individuality except for the fact that identical twins have
side. Iris develops during prenatal period and stabilizes identical DNA patterns. It is, however, currently used mostly
during the first two years of life. The complex iris texture in the context of forensic applications for person recognition.
carries very distinctive information useful for personal Three issues limit the utility of this biometrics for other
recognition. Irises of twins are different as well. Iris based applications: 1) contamination and sensitivity: it is easy to
recognition systems provide promising speed and accuracy steal a piece of DNA from an unsuspecting subject that can be
and support large scale identification operations as well. subsequently abused for an ulterior purpose; 2) automatic
Contact lenses printed with fake iris can be detected. The real-time recognition issues: the present technology for DNA
hippus movement of the eye can also be used for liveness matching requires cumbersome chemical methods (wet
detection. processes) involving an experts skills and is not geared for
The retinal vasculature is rich in structure and is supposed to on-line noninvasive recognition; and 3) privacy issues:
be a characteristic of each individual and each eye. It is information about susceptibilities of a person to certain
claimed to be the most secure biometric since it is not easy to diseases could be gained from the DNA pattern and there is a
change or replicate the retinal vasculature. The image concern that the unintended abuse of genetic code information
acquisition requires a person to peep into an eye-piece and may result in discrimination, e.g., in hiring practices.
focus on a specific spot in the visual field so that a

79 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-2, Issue-8, August 2014
Voice:
Voice is a combination of physiological and behavioral
biometrics. Fig (h) shows how it can be used for
identification. The features of an individuals voice are based
on the shape and size of the appendages (e.g., vocal tracts,
mouth, nasal cavities, and lips) that are used in the synthesis
of the sound. These physiological characteristics of human
speech are invariant for an individual, but the behavioral part
of the speech of a person changes over time due to age,
medical conditions (such as a common cold), and emotional
state, etc. Voice is also not very distinctive and may not be
Fig (e) appropriate for large-scale identification. A text-dependent
voice recognition system is based on the utterance of a fixed
Gait: predetermined phrase. A text-independent voice recognition
As shown in the Fig (f) gait is the peculiar way one walks and system recognizes the speaker independent of what she
is a complex spatio-temporal biometric. Gait is not supposed speaks. A text-independent system is more difficult to design
to be very distinctive, but is sufficiently discriminatory to than a text-dependent system but offers more protection
allow verification in some low-security applications. Gait is a against fraud. A disadvantage of voice-based recognition is
behavioral biometric and may not remain invariant, especially that speech features are sensitive to a number of factors such
over a long period of time, due to fluctuations in body weight, as background noise. Speaker recognition is most appropriate
major injuries involving joints or brain, or due to inebriety. in phone-based applications but the voice signal over phone is
Acquisition of gait is similar to acquiring a facial picture and, typically degraded in quality by the microphone and the
hence, may be an acceptable biometric. Since gait-based communication channel.
systems use the video-sequence footage of a walking person
to measure several different movements of each articulate
joint, it is input intensive and computationally expensive.

Fig (h)

Hand Geometry:
Hand geometry recognition systems, as shown in the Fig (i)
Fig (f) are based on the different measurements such as shape of the
hand, size of palm, lengths and widths of the fingers. Hand
Signature: features are not very distinctive. They are suitable for
The way a person signs his or her name is known to be a verification but not for identification. In certain situations
characteristic of that individual, as shown in the Fig (g). such as immigration and border control, biometrics such as
Although signatures require contact with the writing fingerprints may not be suitable because they infringe on
instrument and an effort on the part of the user, they have been privacy. In such situations hand geometry can be used for
accepted in government, legal, and commercial transactions verification as hand geometry is not very distinctive. Hand
as a method of verification. Signatures are a behavioral geometry features may not be invariant during the growth
biometric that change over a period of time and are influenced period of children. The size of such recognition systems is
by physical and emotional conditions of the signatories. large and hence it is difficult to embed the systems in other
Signatures of some people vary substantially: even successive devices such as laptops.
impressions of their signature are significantly different.
Further, professional forgers may be able to reproduce
signatures that fool the system.

Fig (i)
Fig (g)
A brief comparison of the above biometric techniques based
on seven factors is provided in Table-I. The applicability of a

80 www.erpublication.org
 A Multibiometric Recognition System : Categorization, Retrieval Techniques & Applications

specific biometric technique depends heavily on the modality. Maximum benefit would be derived from
requirements of the application domain. No single technique algorithms that are based on distinctly different and
can outperform all the others in all operational environments. independent principles.
In this sense, each biometric technique is admissible and there
is no optimal biometric characteristic. For example, it is well
known that both the fingerprint-based and iris-based
techniques are more accurate than the voice-based technique.
However, in a tele-banking application, the voice-based
technique may be preferred since it can be integrated
seamlessly into the existing telephone system.

Fig. 1 : Categories of Multibiometric System

Multiinstance biometric systems use one sensor (or possibly

TABLE-I multiple sensors) to capture samples of two or more different
instances of the same biometric characteristics. For example,
systems capturing images from multiple fingers are
II. CATEGORIZATION considered to be multiinstance rather than multimodal.
However, systems capturing, for example, sequential frames
The International Committee for Information Technology of facial or iris images are considered to be multi presentation
Standards (INCITS) Technical Committee M1, Biometrics, rather than multiinstance. This is whether or not the repeated
and researchers have described methods for performing captured images are combined at the image (feature) level,
multibiometric fusion. In general, the use of the terms some other level of combination or a single image is selected
multimodal or multibiometric indicates the presence and use as the one best used for pattern matching.
of more than one biometric aspect (modality, sensor, instance
and/or algorithm) in some form of combined use for making a Multisensorial biometric systems sample the same instance
specific biometric verification/identification decision. The of a biometric trait with two or more distinctly different
goal of multi-biometrics is to reduce one or more of the sensors. Processing of the multiple samples can be done with
following: one algorithm or some combination of multiple algorithms.
False accept rate (FAR) For example, a face recognition application could use both a
False reject rate (FRR) visible light camera and an infrared camera coupled with
Failure to enroll rate (FTE) specific frequency (or several frequencies) of infrared
Susceptibility to artifacts or mimics illumination. For a specific application in an operational
To further the understanding of the distinction among the environment, there are numerous system design
multi-biometric categories, they are briefly summarized in the considerations, and trade-offs that must be made among
following: factors such as improved performance (e.g., verification or
identification accuracy, system speed and throughput,
Multimodal biometric systems take input from single or robustness, and resource requirements), acceptability,
multiple sensors measuring two or more different modalities circumvention, ease of use, operational cost, environment
of biometric characteristics. For example, a system combining flexibility and population flexibility. Especially for a
face and iris characteristics for biometric recognition would large-scale identification system, there are additional system
be design considerations such as operation and maintenance,
considered a multimodal system regardless of whether face reliability, system acquisition cost, life cycle cost and planned
and iris images were captured by different or same imaging system response to identified susceptible means of attacks, all
devices. It is not required that the various measures be of which will affect the overall deploy ability of the system.
mathematically combined in anyway. For example, a system
with fingerprint and face recognition would be considered III. RETRIEVAL TECHNIQUES
multimodal even if the OR rule was being applied,
First Approach :
allowing users to be verified using either of the modalities.
In a biometric identification system, the identity
corresponding to the input data (probe/investigation) is
Multialgorithmic biometric systems take a single sample
typically determined by comparing it against the templates of
from a single sensor and process that sample with two or more
all identities in a database (gallery).Exhaustive/in-depth
different algorithms. The technique could be applied to any

81 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-2, Issue-8, August 2014
matching against a large number of identities increases the matched with the set of reference images already stored in the
response time of the system and may also reduce the accuracy database. So as compared to the number of reference images
of identification. One way to reduce the response time is by the index codes varies the reference images may be viewed as
designing biometric templates that allow for rapid matching. basis vectors in the original feature space. If two images
An alternative approach is to limit the number of identities ,the input and reference images are similar then their values
against which matching is performed based on criteria that are are expected to be lesser than the threshold ,where the
fast to evaluate. In this technique the search space is reduced threshold is set before processing ,if the C values are greater
by partitioning the database into several bins. Following such than threshold
binning, the biometric database will be partitioned such that it is assumed that the two images belongs to different
the templates in each bin are similar and correspond to some individuals. During identification, the indexing system first
natural or statistical class. In case of the traditional 1: N computes the index code S of the probe. Then it outputs all
comparisons for identification, the time needed for the system enrolled identities whose index codes are within a certain
would be to determine the distance between the test template distance from S. The index codes are generated from both
and the N templates in database. Thus the total time needed in face and Fingerprint Images.
such a case could be given as: Q (N). This is achieved by
reducing the search space using Gittins index algorithm and it 3. Input Image Clipping Processing:
also improves the accuracy of identification. Clipping refers to any procedure which identifies the portion
of a picture which is either inside or outside a region using any
clipping algorithm. The region against which an object is to
be clipped is; called clipping window. Clipping is a process of
capturing or processing an image where the intensity in a
certain area falls outside the minimum and maximum intensity
which can be represented. In the input image clipping process,
the input image is partitioned into several patches and each
patch will search for corresponding matches in database. If
any match is found then RGB and color code value is
generated for that image.

4. Selecting the reference Image and retrieving the Image:

Reference images can be selected from the database itself.
They can also be synthetically generated images. While the
entire database can be viewed as a candidate pool for
selecting reference images, practical considerations dictate
the use of a small random subset of images for this purpose.

Second Approach:

The retrieval of a small number of candidate identities from a

database based on the probe data is known as database
filtering. Filtering can be accomplished by using
classification or indexing schemes. In a classification scheme,
identities in the database are partitioned into several classes.
Only the identities belonging to the same class as that of the
probe image are retrieved during the search process for
Fig. 2: Architecture Diagram further comparison. This approach has two main limitations:
1) it assumes that each identity can be unambiguously
1. Dataset Pre-processing: assigned to a single class; and 2) the distribution of identities
In the dataset preprocessing the images of face and finger across classes may be uneven resulting in inefficient
print are collected and stored in the database .For face images classification.
we use dataset from FERET and the FRGC and fro finger
print images we use WVU fingerprint database. There are In contrast, the goal of an indexing scheme is to assign a
1195 subjects with frontal face images in the FERET unique index value to every identity in the database. However,
database. We use only 1010 of these subjects because the the index value of the probe image may not be identical to that
images of the remaining 185 subjects could not be processed of the corresponding identity in the database because the
The WVU fingerprint database contains images of 4 different process of biometric acquisition and processing is susceptible
fingers (left index, left thumb, right index, right thumb) from to noise. Therefore, the retrieval scheme has to employ some
270 subjects. We treated the individual fingers as independent type of neighborhood search in the index space. An efficient
subjects, resulting in a total of 1080 subjects. indexing algorithm retrieves a small number of candidate
identities based on similarity measures that can be computed
2. Generation of Index codes for all the Images: quickly. An important advantage of indexing techniques is
The index code of an image is the list of its match scores that they do not create boundaries among the continuously
against the reference images. An image is taken and it is distributed templates.

82 www.erpublication.org
 A Multibiometric Recognition System : Categorization, Retrieval Techniques & Applications

Fig. 6: Fusion by union of candidate lists

A. Concatenation of Index Codes

Let Sx (Ri) = {s (xi, r1i), s (xi, r2i) ,, s (xi, rni)} be the index
Fig. 3: Generation of an index code. An input image is code of identity , where i denotes the modality, rj denotes the
matched against a set of reference images. The set of resulting jth reference image in this modality, and s(a, b) denotes the
match scores constitutes the index code of this input image. match score between a and b. The fusd index codes from
different modalities:
Index Codes For Multimodal Databases Fx = { s (x1, r11),, s (x1, rn1), s (x2, r12),, s (x2, rn2)}.
There is an inherent trade-off between the total number of
retrieved candidates and the number of correctly retrieved Fig. 5 illustrates this process schematically. Retrieval using
candidates. Fusion schemes are often useful for narrowing the fused index code is performed as for a single modality.
down the total number of retrieved candidates and/or This fusion scheme results in longer index codes. Ideally,
increasing the number of correctly retrieved candidates. In using longer index codes results in larger variances among
biometric identification, it is crucial that the correct identity is themthis is desirable. One weakness of this fusion scheme
in the candidate list even if this results in a longer list. We is that poor indexing performance due to one of the modalities
propose two fusion techniques that use the information from can negatively affect the overall performance of indexing.
multiple modalities in a complementary manner. Index codes
are stored separately for each modality thereby making the B. Union of Candidate Lists
indexing scheme flexible in including more modalities or Another fusion mechanism is to combine the lists of candidate
excluding a certain modality. The ability to exclude a identities output by each modality. Let Ci be the set of
modality from the indexing process is valuable when prior retrieved identities according to modality. The final set of
knowledge indicates that a certain modality is unreliable or identities retrieved by the indexing will be C = Ui=1k Ci as
when data for a modality are missing. This approach for shown in Fig. 6. This fusion scheme has the potential to
indexing multimodal databases is shown in Fig. 4. increase the chances of finding the right identity in even if the
right identity is not located in some of the Cis. Thus, poor
indexing performance of one modality would have a smaller
effect on the overall indexing performance. This approach
fails only when the right identity is not retrieved by any of the
modalities. Intersection of the identities in the candidate lists
is another option for indexing multimodal databases but is not
discussed in this paper due to its inferior performance.

IV. APPLICATIONS

Applications can be categorized into three main groups:

1) Commercial applications such as computer network login,

e-commerce, Internet access, ATMs or credit cards, physical
access control, mobile phones, Personal Digital Assistant
Fig. 4: Indexing two modalities. Two index codes are (PDA)s, medical records management, distance learning, etc.
generated separately, one for each modality. The information
from the two modalities is combined during retrieval. 2) Government applications such as national ID card, drivers
license, social security, border control, passport control,
welfare-disbursement, etc.

83 www.erpublication.org
 International Journal of Engineering and Technical Research (IJETR)
ISSN: 2321-0869, Volume-2, Issue-8, August 2014

3) Forensic applications such as corpse identification,

criminal investigation, terrorist identification, parenthood
determination, etc.

V. CONCLUSION
This paper overviews and discusses the various scenarios that
are possible in multimodal biometric systems using
fingerprint, face and iris recognition, the techniques that can
be adopted to retrieve the information and improve overall
system accuracy.

REFERENCES
[1] A. Ross and A.K. Jain, Information Fusion in Biometrics, Pattern
Recognition Letter 24, pp.2115-2125, 2003.
[2] A. Jain, R. Bolle, and S. Pankanti, editors, Biometrics: Personal
Identification in Networked Society, Kluwer Academic Publishers,
1999.
[3] ISO/IEC JTC 1/SC 37 Biometrics, Working Draft Technical Report on
Multi-Modal and Other Multi-Biometric Fusion, August 2005.
[4] S. Prabhakar and A.K. Jain, Decision-level fusion in fingerprint
verification, Pattern Recognition, Vol. 35, No. 4, pp. 861-874, 2002.
[5] H. Korves, L. Nadel, B. Ulery, and D. Masi, Multibiometric Fusion:
From Research to Operations, Sigma, Mitretek Systems, Summer
2005.
[6] Summary of NIST Standards for Biometric Accuracy,
Tamper Resistance, and Interoperability, November 13,
2002.
[7] A. Mhatre, S. Palla, S. Chikkerur, and V. Govindaraju, Efficient search
and retrieval in biometric databases, Biometric Technol.Human
Identification II, vol. 5779, no. 1, pp. 265273, 2005
[8] A. Gyaourova and A. Ross, A coding scheme for indexing multimodal
biometric databases, in Proc. IEEE Computer Society Workshop on
Biometrics at the Computer Vision and Pattern Recognition (CVPR)
Conf., Miami, FL, Jun. 2009.
[9]http://www.biometric solutions.com/applications/index.php
[10] http://www.intechopen.com/books/recent-application-in-biometrics

84 www.erpublication.org