Preimage Attack on BioHashing

Patrick Lacharme, Estelle Cherrier and Christophe Rosenberger

Normandie Univ., F-14032 Caen, France

UNICAEN, GREYC, F-14032 Caen, France

ENSICAEN, GREYC, F-14032 Caen, France

CNRS, UMR 6072, F-14032 Caen, France

Keywords:

Biometrics, Security, Privacy, Data Protection, Cancelable Biometrics, BioHashing, Spooﬁng Attacks,

Genetic Algorithms.

Abstract:

Biometric recognition is more and more employed in authentication and access control of various applica-

tions. Biometric data are strongly linked with the user and do not allow revocability nor diversity, without an

adapted post-processing. Cancelable biometrics, including the very popular algorithm BioHashing, is used to

cope with the underlying privacy and security issues. The principle is to transform a biometric template in

a BioCode, in order to enhance user privacy and application security. These schemes are used for template

protection of several biometric modalities, as ﬁngerprints or face and the robustness is generally related to the

hardness to recover the original biometric template by an impostor. In this paper, we propose to use genetic al-

gorithms to approximate the original biometric feature and spoof the authentication system. We show through

experimental results on ﬁngerprints the efﬁciency of the proposed attack on the BioHashing algorithm, by

approximating the original FingerCode, given the seed and the corresponding BioCode.

1 INTRODUCTION

Biometrics is a major concern for privacy as it has a

direct link with the user and is generally non revoca-

ble, without an adapted post-processing. Indeed, if a

biometric data is stolen or compromised, it is difﬁcult

(if not impossible) to revoke it, contrary to a classical

password. Moreover, the same biometric data may be

used for several applications, resulting in important

threats for the security, in the absence of a strong di-

versiﬁcation process. In the same time, biometrics is

more and more deployed in various applications, as

in electronic passport, access control, electronic pay-

ment or forensics applications. Common vulnerabili-

ties of biometric schemes include spooﬁng and replay

attacks or collection of biometric data without the

consent of people. As a consequence, biometrics de-

velopment provides an important technological chal-

lenge for data security and user privacy (Cavoukian

and Stoianov, 2009).

Biometric template protection schemes are a

group of technologies, included in privacy enhancing

technologies, used to enhance both privacy and secu-

rity of biometric data. Therefore, any template protec-

tion approach should allow the possibility to revoke

a biometric data in case of interception, and should

be carefully designed, with a strong security analy-

sis. Among the different solutions in the literature,

template protection can be achieved using biometric

cryptosystems or by cancelable biometrics (Rathgeb

and Uhl, 2011). These biometric template protection

schemes strongly depend on the biometric modalities.

For example an Iriscode, encoded as a binary vector

of 256 bytes could not be protected in the same way

than a set of minutiae of varying lengths. Some of

these schemes have been recently normalized in the

standard ISO 24745 (ISO, 2011).

The concept of cancelable biometrics relies on a

transformation of the raw biometric data, enabling the

transformed data to address security and privacy pro-

tection issues. The general principle consists in the

generation of a new biometric template, from the bio-

metric feature vector (such as texture parameters) and

a random number. Therefore, it can be seen as a two-

factor authentication scheme. At the enrolment stage,

once transformed, the new template (or transformed

template) is stored in the database, while the original

raw biometric vector is discarded and never kept. At

the veriﬁcation stage a comparison is performed be-

tween two transformed templates: between the one

the user pretends to correspond and between the one

he/she presents to the system. Hence, to be authen-

363

Lacharme P., Cherrier E. and Rosenberger C..

Preimage Attack on BioHashing.

DOI: 10.5220/0004524103630370

In Proceedings of the 10th International Conference on Security and Cryptography (SECRYPT-2013), pages 363-370

ISBN: 978-989-8565-73-0

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

ticated, a user must present the same biometric data

(more precisely a similar biometric data) and the same

random number. Howeverthis random number is gen-

erally not considered as secret, in the sense that it is

generally stored with the transformed template for the

veriﬁcation step. Cancelable biometric systems must

meet the following four criteria, (Maltoni et al., 2003;

Jain et al., 2008; Nagar et al., 2010):

• Performance

The template transformation should not signiﬁ-

cantly decrease the technical performance of the

original biometric system (accuracy).

• Revocability

It should be possible to revoke a biometric tem-

plate in case of compromission, and to generate a

new one from the original data.

• Irreversibility

From the transformed data, it should not be possi-

ble to obtain enough information about the origi-

nal biometric data.

• Unlinkability

It should be possible to generate different trans-

formed data for multiple applications, and no in-

formation should be deduced from the compari-

son or the correlation of different realizations.

The advantage of cancelable biometrics lies in the

ease of revoking the transformed template, by sim-

ply changing the associated random number. An-

other interest lies in the possibility to generate dif-

ferent templates to authenticate oneself to different

services from the same biometric raw data, with dis-

tinct random numbers (one for each service). Thus,

the random number should only be used for diver-

siﬁcation and revocability purposes and not for the

security, without a secure storage. More precisely,

the security of any cancelable biometric process re-

quires the associated transformation to be non invert-

ible: it means that it should be hard for an intruder

to recover the original raw biometric vector from the

transformed template and the random number. No-

tice that with this commonly admitted deﬁnition of

the non-invertibility property, the possibility to ap-

proximate the original biometric feature vector, given

the transformed template and the associated random

number, is not considered. Nevertheless, the recon-

struction of such sufﬁciently similar biometric tem-

plates, called preimage attack, is a major ﬂaw for

cancelable biometric schemes, because in this case,

the authentication system could be spoofed. Different

deﬁnition for irreversibility are detailed in (Simoens

et al., 2012) with several criteria: full-leakage ir-

reversibility, authorized-leakage irreversibility and

pseudo-authorized leakage irreversibility. For exam-

ple, it is possible to generate an eligible ﬁngerprint

given minutiae (Cappelli et al., 2007).

The BioHashing algorithm is one of the most pop-

ular cancelable biometric scheme, proposed for face

biometrics in (Goh and Ngo, 2003) and later for ﬁn-

gerprints in (Teoh et al., 2004), which will be detailed

hereafter. The invertibility of the Biohashing algo-

rithm has been ﬁrstly investigated in (Cheung et al.,

2005; Lee et al., 2009). Recently, Nagar et al. pre-

sented a method based on optimization problems, to

recover a close approximation of face images, gener-

ated by the Biohashing algorithm (Nagar et al., 2010).

The main contribution of this paper is to analyze

this vulnerability of cancelable biometrics. We pro-

pose a new method to generate a biometric feature

vector approximating the original biometric feature,

based on genetic algorithms. Experiments are carried

out on ﬁngerprint modality, with the BioHashing al-

gorithm, using the FVC2002 benchmark.

This paper is organized as follows. Section 2 pro-

vides a presentation of cancelable biometric schemes,

with a description of the BioHashing algorithm. Sec-

tion 3 then introduces genetic algorithms and their

application to template approximation. Finally, Sec-

tion 4 proposes experimental results on the FVC2002

database with the BioHashing algorithm.

2 BIOMETRIC DATA

PROTECTION

Biometric systems are used for identiﬁcation or au-

thentication purpose. Identiﬁcation process generally

involves a large database of biometric templates and

the veriﬁcation phase consists in recovering the cor-

responding template in the database. The centralized

storage of non-protected biometric data is a major

threat for user privacy. Biometric authentication does

not necessarily use a centralized database and many

applications require an additional secure element as a

smart card for biometric data storage. However, the

centralized storage of protected biometric data is a

possible alternative, if this centralized approach is not

a privacy nor a security threat for the system.

2.1 Biometric Cryptosystems

Biometric cryptosystems associate a secret key with a

biometric template in order to protect the latter. It

includes fuzzy commitment, (Juels and Wattenberg,

1999) and fuzzy vaults, (Juels and Sudan, 2002).

Fuzzy commitments are based on error correcting

codes and do not require the storage of the biomet-

ric template. They have numerous applications on

SECRYPT2013-InternationalConferenceonSecurityandCryptography

364

Figure 1: Architecture for identity management using cancelable biometrics.

iris data (Hao et al., 2005) or multimodal systems

(Cimato et al., 2008). Fuzzy vaults are especially

suitable for partial biometric representations, as for a

minutiae representation of ﬁngerprints (Nandakumar

et al., 2007;

Orencik et al., 2008). These schemes

have been formalized in fuzzy sketches and fuzzy ex-

tractors in order to derive cryptographic keys from

noisy biometric data (Dodis et al., 2004; Boyen,

2004). Fuzzy commitments are suitable for bio-

metric data having a binary representation, like the

Iriscode (Daugman, 2004; Daugman, 2007). How-

ever, several weaknesses are presented in (Simoens

et al., 2009; Blanton and Aliasgari, 2011; Zhou et al.,

2012). In the same way, collusion attacks on the

fuzzy vault scheme are proposed in (Schreier and

Boult, 2007; Poon and Miri, 2009). Finally biomet-

ric cryptosystems combined with private information

retrieval (PIR) protocols (Bringer et al., 2007) or ho-

momorphic encryption have recently given interesting

solutions for face biometrics (Osadchy et al., 2010),

iris and ﬁngerprints (Barni et al., 2010; Blanton and

Gasti, 2011).

2.2 Cancelable Biometrics and

BioHashing

Cancelable biometric systems have been designed to

ensure the privacy of the use of biometric data. The

feature transformations were ﬁrst proposed in (Ratha

et al., 2001; Bolle et al., 2002) and many cancelable

biometric schemes have been proposed later. In ad-

dition to the BioHashing algorithm, we can mention

the approach proposed in (Ratha et al., 2007), where

the authors use three geometric transformations to be

applied to minutiae: Cartesian, Polar and Functional

transformations. The centralized storage of Biocodes

is not a security problem if data are revokable and

if the transformation is non-invertible. In this case,

the storage of the additional random number must

be carefully handled. This data could be stored in a

secure element for each service provider, but an al-

ternative solution seems possible, where the identity

provider stores for any user identiﬁer (i) a BioCode

and (ii) the seed value for each associated service

provider. Figure 1 describes this alternative.

In biometric feature transformation schemes, the

biometric feature vector is generally represented by

a real-valued vector and the metric used to evalu-

ate the similarity between two biometric features is

the Euclidean distance. Technical performance of the

system (accuracy and accuracy degradation caused

by the template protection scheme) is generally mea-

sured with FMR/FNMR or FAR/FRR rates (respec-

tively False match rate, false non-match rate, false ac-

ceptation rate and false reject rate). Feature transfor-

mations should clearly preserve the performance of

the biometric system, according to the aformentioned

properties. Among the papers dealing with cance-

lable biometrics, most of them rely on the BioHash-

ing scheme, since the algorithm is easy to analyse and

can be used on several biometric modalities. More-

over, the multiplication with the orthogonal matrix

preserves the scalar product (more details are given

in Section 2.3) and consequently the technical perfor-

mance of the system.

At the enrolment step, the original biometric fea-

ture vector, called FingerCode (on account of the cho-

sen ﬁngerprint modality), is transformed using a ran-

dom number (called the seed), into a new template,

called the reference BioCode. Once the transforma-

tion is achieved, the original FingerCode is discarded

and the reference BioCode is stored, with the asso-

ciated seed. At the veriﬁcation step, a new BioCode

is computed using the same algorithm, with a second

PreimageAttackonBioHashing

365

biometric vector and the corresponding random seed.

The veriﬁcation result is obtained from the computa-

tion of a simple Hamming distance between the refer-

ence BioCode and the one issued from the new cap-

ture. Figure 2 illustrates the overall process, with the

BioHashing scheme applied to ﬁngerprints.

Figure 2: General principle of BioHashing on ﬁngerprints.

It is known that the Biohashing algorithm is easy

to invert if the random number used as the seed

is known. But, the reconstructed biometric feature

(called the preimage) is not necessary close to the

original template. In this paper, we wonder whether

it may be sufﬁcient for an intruder to retrieve, from

the knowledge of an intercepted BioCode and the cor-

responding random number, an approximated Finger-

Code. We recall that the FingerCode is the vector con-

taining featuresextracted from the original raw ﬁnger-

print data of the user, and not the original data itself.

Nagar et al. have recently presented the ﬁrst de-

tailed method to recover a close approximation of the

original face image given the BioCode and the ran-

dom seed (Nagar et al., 2010). Nevertheless, the aim

for the intruder is to be accepted by the system thanks

to the approximate FingerCode and not necessary to

retrieve an approximate ﬁngerprint. More precisely,

with the knowledge of a BioCode and the associated

seed, we investigatethe possibility to approximate the

corresponding FingerCode and the possibility to gen-

erate other BioCodes with this approximated Finger-

Code.

2.3 The BioHashing Algorithm

In this section, we give some details about the Bio-

Hashing algorithm. This algorithm transforms a real-

valued vector of length n (i.e. the FingerCode, re-

sulting from a feature extraction method) into a bi-

nary vector of length n (i.e. the BioCode), as ﬁrst

deﬁned by Teoh et al. in (Teoh et al., 2004). The Bio-

Hashing algorithm is mainly used for ﬁngerprints and

face modalities. Its principle consists in projecting

the FingerCode on an orthogonal basis deﬁned by the

random seed, to generate the BioCode. The template

transformation uses the following algorithm, where

the inputs are the random seed and the FingerCode

F and the output is the BioCode B:

1. For i = 1, . . . ,n, n pseudorandom vectors v

length n are generated (from the random seed) and

are gathered in a pseudorandom matrix.

2. The Gram-Schmidt algorithm is applied on the n

vectors v

of the matrix, for the generation of n

orthonormal vectorsV

, . . . ,V

3. For i = 1, . . . , n, n scalar products p

=< F,V

are computed using the FingerCode F and the n

orthonormal vectorsV

4. The n-bit biocode B = (B

, . . . , B

) is ﬁnally ob-

tained, using the following quantization process:



0 if p

< t

1 if p

≥ t,

where t is a given threshold, generally equal to 0.

Note that the distance between two FingerCodes is

computed with the Euclidean distance, while between

two BioCodes, the Hamming distance is used.

The Gram-Schmidt algorithm transforms an arbi-

trary basis into an orthogonal basis. Roughly speak-

ing, the ﬁrst part of the algorithm, including the scalar

products with the orthonormal vectors, is used for the

performance requirements and the last step of the al-

gorithm is used for the non-invertibility requirements

of the BioHashing algorithm. As mentioned before,

the random seed guarantees the diversity and revoca-

bility properties.

3 FINGERCODE

APPROXIMATION WITH

GENETIC ALGORITHMS

In this section, we present how to obtain an approx-

imate FingerCode, from an intercepted BioCode and

the corresponding random seed, resorting to genetic

algorithms.

3.1 Introduction to Genetic Algorithms

A genetic algorithm is any population-based model

that uses selection and recombination operators to

generate new sample points in a search space (Whit-

ley, 1994). A genetic algorithm simulates computa-

tional models inspired from evolutionary theory, us-

ing the following principle: a random initial popu-

lation is generated. A criterion is deﬁned, typically

a ﬁtness function, in the sense that this criterion is

evaluated for each individual, to enable a comparison

and a selection between the different individuals. The

individuals obtaining the best score with respect to

the criterion are kept, while the others are discarded.

SECRYPT2013-InternationalConferenceonSecurityandCryptography

366

Then, inspired from the evolutionary theories, some

process of cross-over and mutation are applied, to ob-

tain a new generation of individuals from the previous

one. The processes of cross-over and mutation avoid

the algorithm to fall into local extrema.

More precisely, genetic algorithms (or GA) deter-

mine the optimal value of a criterion by simulating the

evolution of a population and survival of best ﬁtted in-

dividuals (Wall, 1996). The survivors are individuals

obtained by crossing-over, mutation and selection of

individuals from the previous generation. GA is an

optimization method that does not necessitate to dif-

ferentiate the ﬁtness function but only to evaluate it.

If the population is important enough considering the

size of the search space, the ﬁtness criterion is guar-

anteed to reach its optimal value.

3.2 Application to Cancelable

Biometrics

We use the following notations, introduced in (Nagar

et al., 2010). Let b

and

represent the template and

query biometric features (or FingerCodes) of user z,

respectively. Let f be the feature transformation func-

tion (i.e. the orthonormal projection followed by the

quantization) and K

be the random seed (or transfor-

mation parameters) corresponding to the user z. The

resulting enroled BioCode is denoted B

= f(b

, K

)

and n is the dimension of the BioCode. In this sec-

tion, we use a genetic algorithm to approximate b

knowing B

and the secret data K

. We use the termi-

nology of the Biohashing algorithm (FingerCode and

BioCode), but all this section is directly applicable to

any feature transformation f. Our approach uses ge-

netic algorithms and -to our knowledge- it is the ﬁrst

time that such algorithms are used in biometrics for

this purpose.

A genetic algorithm is deﬁned by considering ﬁve

essential data, applied here to our FingerCode approx-

imation problem :

1. Genotype: a candidate FingerCode denoted

considered as an individual described by a vector

of dimension m,

2. Population: a set composed of 10.000 individu-

als characterized by their genotypes (i.e. a set

of 10.000 candidates for the approximation of the

FingerCode),

3. Fitness Function: this function enables to quan-

tify the ﬁtness of an individual to the environ-

ment by considering its genotype. Considering

the problem of FingerCode approximation, we

propose to use the following ﬁtness function:

) =



, K

) − B



(1)

It means that the intruder wants to retrieve a

new FingerCode

which is an approximation of

the original one b

, from the knowledge of the

Biocode B

= f(b

, K

) and the associated random

seed K

4. Operators on Genotypes: they deﬁne alterations

on genotypes in order to make the population

evolve during generations. Three types of oper-

ators are used:

• Mutation step: individual’s genes are modiﬁed

in order to be better adapted to the environ-

ment. We use the non-uniform mutation pro-

cess which randomly selects one chromosome

, and sets it as equal to a non-uniform random

number:

′



+ (b

− x

)h(G) if r

< 0.5

− (x

+ a

)h(G) if r

≥ 0.5

(2)

where

h(G) = (r

(1−

max

))

, r

: numbers belonging to the interval [0, 1]

, b

: lower and upper bound of chromosome x

G : the current generation

max

: the maximum number of generations

b : a shape parameter

(3)

• Selection step: individuals that are not adapted

to the environment do not survive to the next

generation. We used the normalized geometric

ranking selection method which deﬁnes a prob-

ability P

for each individual i to be selected as

following:

q(1− q)

r−1

1− (1− q)

(4)

where

q : the probability of selecting the best

individual

r : the rank of individual, where 1 is the best

n : the size of the population

(5)

• Crossing-over step: two individuals can repro-

duce by combining their genes. We use the

arithmetic crossover which produces two com-

plementary linear combinations of the parents:

′

= aX + (1− a)Y

′

= (1− a)X + aY

(6)

where

X,Y : genotype of parents

a : a number in the interval [0, 1]

′

: genotype of the linear combinations

of the parents

(7)

PreimageAttackonBioHashing

367

5. Stopping criterion : this criterion allows to stop

the evolution of the population. We can consider

the stability of the standard deviation of the evalu-

ation criterion of the population or set a maximal

number of iterations (we used the second one with

the number of iterations equal to 2000).

The implementation of this algorithm is presented

in the next section with the BioHashing algorithm on

a ﬁngerprints database.

4 EXPERIMENTAL RESULTS

We generated a FingerCode for each ﬁngerprint in

the FVC2002 database (Maio et al., 2002) dB3, com-

posed of 8 ﬁngerprints by individual (resolution 355

x 390 pixels) for 100 individuals. The FingerCode of

each user is generated following two feature compu-

tation methods (providing different dimensions of the

vector):

• Method 1 : Gabor features (Manjunath and Ma,

1996) with 256 parameters. They are based on

a Gaussian kernel function modulated by a sinu-

soidal plane wave, with several different orienta-

tions and scales, and are used for texture represen-

tation.

• Method 2 : Rotation invariant local binary pattern

(LBPFT) with 152 parameters. This is a rotation

invariant texture classiﬁcation method, presented

in the reference (Guo et al., 2010).

For each FingerCode, we computed one BioCode

with a random seed, using the BioHashing algorithm.

We apply the algorithm described in section 3.2 to ap-

proximate the value of the FingerCode, given the as-

sociated BioCode and seed. Figure 3 explains in an

intuitive way the BioHashing process, it consists in

projecting the FingerCode on a unit sphere.

In the validation process, we intend to show how

the proposed method is able to approximate the Fin-

gerCode considering the ﬁtness function (distance be-

tween the real BioCode and the predicted one). Sec-

ond, we have to verify that the predicted FingerCode

can be reused. To do that, given one FingerCode in

the database, we generate two BioCodes (with differ-

ents values of K

). The ﬁrst BioCode and the asso-

ciated parameter K

are used to approximate the Fin-

gerCode. The second BioCode is used to verify if

the predicted FingerCode provides a similar BioCode

(i.e. if



f(b

, K

′

) − f(

, K

′

)



is small).

Figure 3: BioCode computation and FingerCode approxi-

mation.

4.1 FingerCode ASpproximation

We applied the general process described in section

3.2 on FingerCodes generated with method 1 and

method 2. The evolution of the ﬁtness function for the

the LBPFT (dimension 152) and Gabor (dimension

256) features is presented in ﬁgure 4. The average

optimal value of the ﬁtness function (1) equals 5.5 for

the LBPFT feature extraction method and 9.2 for the

Gabor one. This means that the generated BioCode,

given the approximated FingerCode, is very similar

to the intercepted one. The resulting average differ-

ence between the real FingerCode and the approxi-

mated one was 6.4 for the LBPFT feature (dimension

152) and 203.2 for the Gabor one (dimension 256). It

shows the efﬁciency of the proposed approach for the

LBPFT feature. For the Gabor one, the distance be-

tween the real and approximated FingerCode is quite

high (but computed on dimension 256).

4.2 Reusability of the Approximated

FingerCode

In order to verify if the approximated FingerCode is

really useful for an attacker, we undertook another ex-

periment. The hypotheses are: the attacker has stolen

the random seed and has generated an approximate

FingerCode. Therefore, we generated two BioCodes

with a new random seed (i) with the original Finger-

Code and (ii) with the approximatedone. From the at-

tacker’s viewpoint, we expect to obtain two BioCodes

that are similar enough (sufﬁciently to be accepted by

the cancelable biometric system). We obtained an av-

erage distance between the BioCodes equal to 8.6 for

the LBPFT method and 11.0 for the Gabor method.

The conclusion of this experiment is that the pro-

posed attack reveals a real problem of security if the

BioCode and random seed are stored in the same

database. It is also a privacy issue since an intruder

would be able to generate other BioCodes to imper-

sonate a genuine user.

SECRYPT2013-InternationalConferenceonSecurityandCryptography

368

(a)

(b)

Figure 4: Evolution of the ﬁtness function for each genera-

tion: (a) LBPFT, (b) Gabor.

5 CONCLUSIONS AND

PERSPECTIVES

This paper proposes a new way to approximate the

original biometric feature from the transformed tem-

plate in a cancelable biometric scheme. Our approach

uses genetic algorithms and reveals security and pri-

vacy problems concerning the associated cancelable

biometric system. This attack enables an intruder to

recover a biometric template, similar to the original

template, under realistic assumptions. Experimenta-

tions with the BioHashing algorithm clearly show the

importance of storing the additional random data in a

secure element, apart from the Biocode. Perspectives

of this paper are to study the robustness of different

feature transformations to be used in the context of a

cancelable biometric system.

REFERENCES

Barni, M., Bianchi, T., Catalano, D., Raimondo, M. D., La-

bati, R. D., Failla, P., Fiore, D., Lazzeretti, R., Pi-

uri, V., Piva, A., and Scotti, F. (2010). A privacy-

compliant ﬁngerprint recognition system based on ho-

momorphic encryption and ﬁngercode templates. In

IEEE Fourth International Conference On Biomet-

rics: Theory, Applications And Systems (BTAS 2010).

Blanton, M. and Aliasgari, M. (2011). On the (non)

reusability of fuzzy scketches and extractors and secu-

rity in the computational setting. In SECRYPT, pages

68–77.

Blanton, M. and Gasti, P. (2011). Secure and efﬁcient pro-

tocols for iris and ﬁngerprint identiﬁcation. In ES-

ORICS, pages 190–209.

Bolle, R., Connell, J., and Ratha, N. (2002). Biometric per-

ils and patches. Pattern Recognition, 35(12):2727–

2738.

Boyen, X. (2004). Reusable cryptographic fuzzy extractors.

In ACM CCS, pages 82–91.

Bringer, J., Chabanne, H., Pointcheval, D., and Tang, Q.

(2007). Extended private information retrieval and its

application in biometrics authentications. In CANS,

pages 175–193.

Cappelli, R., Lumini, A., Maio, D., and Maltoni, D. (2007).

Fingerprint image reconstruction from standard tem-

plates. IEEE Transactions on Pattern Analysis and

Machine Intelligence, 29(9):1489–1503.

Cavoukian, A. and Stoianov, A. (2009). Biometric encryp-

tion.

Cheung, K. H., Kong, A. W., You, J., and Zhang, D. (2005).

An analysis on invertibility of cancelable biometrics

based on biohashing. In CISST’05, pages 40–45.

Cimato, S., Gamassi, M., Piuri, V., Sassi, R., and Scotti,

F. (2008). Privacy-aware biometrics: Design and im-

plementation of a multimodal veriﬁcation system. In

Proceedings of ACSAC’08, pages 130–139.

Daugman, J. (2004). How iris recognition works. Circuits

and Systems for Video Technology, IEEE Transactions

on, 14(1):21–30.

Daugman, J. (2007). New methods in iris recognition. IEEE

Transactions on Systems, Man, and Cybernetics, Part

B: Cybernetics, 37(5):1167–1175.

Dodis, Y., Reyzin, L., and Smith, A. (2004). Fuzzy extrac-

tors: How to generate strong keys from biometrics and

other noisy data. In EUROCRYPT’04,pages 523–540.

Springer-Verlag.

Goh, A. and Ngo, D. (2003). Computation of cryptographic

keys from face biometrics. In Communications and

Multimedia Security, pages 1–13. LNCS 2828.

Guo, Z., Zhang, L., Zhang, D., and Zhang, S. (2010). Rota-

tion invariant texture classiﬁcation using adaptive lbp

with directional statistical features. In IEEE Interna-

tional Conference on Image Processing (ICIP).

Hao, F., Anderson, R., and Daugman, J. (2005). Combining

cryptography with biometrics effectively. University

of Cambridge Computer Laboratory, Tech. Rep.

ISO (2011). ISO/IEC 24745 information technology - se-

curity techniques -biometric information protection.

PreimageAttackonBioHashing

369

Jain, A. K., Nandakumar, K., and Nagar, A. (2008). Bio-

metric template security. EURASIP J. Advances in

Signal Processing, 8(2):1–17.

Juels, A. and Sudan, M. (2002). A fuzzy vault scheme. In

ISIT, page 408.

Juels, A. and Wattenberg, M. (1999). A fuzzy commitment

scheme. In ACM Conference on Computer and Com-

munications Security, pages 28–36.

Lee, Y., Chung, Y., and Moon, K. (2009). Inverse operation

and preimage attack on biohashing. In Workshop on

Computational Intelligence in Biometrics.

Maio, D., Maltoni, D., Cappelli, R., Wayman, J. L., and

Jain, A. K. (2002). FVC2002: Second ﬁngerprint ver-

iﬁcation competition. In ICPR, pages 811 – 814.

Maltoni, D., Maio, D., Jain, A., and Prabhakar, S. (2003).

Handbook of Fingerprint Recognition. Springer.

Manjunath, B. S. and Ma, W. (1996). Texture features for

browsing and retrieval of image data. IEEE Transac-

tions on Pattern Analysis and Machine Intelligence,

18:37–42.

Nagar, A., Nandakumar, K., and Jain, A. K. (2010). Bio-

metric template transformation: A security analy-

sis. Proceedings of SPIE, Electronic Imaging, Media

Forensics and Security XII.

Nandakumar, K., Jain, A., and Pankanti, S. (2007).

Fingerprint-based fuzzy vault: Implementation and

performance. IEEE Transactions on Information

Forensics and Security.

Orencik, C., Pedersen, T. B., Savas, E., and Keskinoz, M.

(2008). Improved fuzzy vault scheme for ﬁngerprint

veriﬁcation. In SECRYPT, pages 37–43.

Osadchy, M., Pinkas, B., Jarrous, A., and Moskovich, B.

(2010). Sciﬁ - a system for secure face identiﬁcation.

In IEEE Symposium on Security and Privacy.

Poon, H. and Miri, A. (2009). A collusion attack on the

fuzzy vault scheme. ISC International Journal of In-

formation Security, 1(1):27–34.

Ratha, N., Chikkerur, S., Connell, J., and Bolle, R. (2007).

Generating cancelable ﬁngerprint templates. IEEE

Transactions on PAMI, 29(4):561–572.

Ratha, N., Connell, J., and Bolle, R. (2001). Enhancing se-

curity and privacy in biometrics-based authentication

system. IBM Systems J., 37(11):2245–2255.

Rathgeb, C. and Uhl, A. (2011). A survey on biometric

cryptosystems and cancelable biometrics. EURASIP

J. on Information Security, 3.

Schreier, W. and Boult, T. (2007). Cracking fuzzy vaults

and biometric encryption. In Biometrics Symposium.

Simoens, K., Chang, C., and Preneel, B. (2009). Privacy

weaknesses in biometric sketches. In 30th IEEE Sym-

posium on Security and Privacy.

Simoens, K., Yang, B., Zhou, X., Beato, F., Busch, C.,

Newton, E. M., and Preneel, B. (2012). Criteria to-

wards metrics for benchmarking template protection

algorithms. In ICB’12, pages 498–505.

Teoh, A., Ngo, D., and Goh, A. (2004). Biohashing: two

factor authentication featuring ﬁngerprint data and to-

kenised random number. Pattern recognition, 40.

Wall, P. (1996). A Genetic Algorithm for Resource-

Constrained Scheduling. PhD thesis, MIT.

Whitley, D. (1994). A genetic algorithm tutorial. Statistics

and Computing, pages 65–85.

Zhou, X., Kuijper, A., and Busch, C. (2012). Cracking iris

fuzzy commitment. In ICB’12.

SECRYPT2013-InternationalConferenceonSecurityandCryptography

370