Converting Fingerprint Local Features to Public Key Using Fuzzy Extractor Mohammed S. Khalil1 and Dzulkifli Muhammad1, M.Masroor Ahmed2 Department of Computer Graphics and Multimedia, Universiti Teknologi Malaysia,
[email protected] 1 Department of Computer Graphics and Multimedia, Universiti Teknologi Malaysia,
[email protected] 2 Department of Computer Graphics and Multimedia, Universiti Teknologi Malaysia,
[email protected] 1
ABSTRACT Biometric security systems are being widely used for ensuring maximum level of safety. In Biometric system, neither the data is uniformly distributed, nor can it be reproduced precisely. Each time it is processed. However, this processed data cannot be used as a password or as a cryptography secret. This paper proposes a novel method to extract fingerprint minutiae features and converting it to a public key using the fuzzy extractor. The public key can be used as a key in a cryptographic application Keywords: Biometric, Fingerprint, Local feature, Fuzzy extractor, and Cryptography.
1. Introduction Biometric system ensures an automatic identification of human being based on the principle of measureable physiological or behavioral characteristics such as fingerprint, an iris pattern, or a voice sample [1]. A significant feature associated with Biometric data is that: it is hard to forge, unique to each person, and excellent source of entropy which makes them an excellent candidate for security applications. However, they have some disadvantage, such as biometric data cannot be subjected to any change, biometric template easy to steal, the data are not uniformly distributed and exactly reproducible, they cannot be
stolen biometric template, and stolen biometric data are stolen for life [3]. There have been several researches in the literature addressing this issue [4; 5; 6; 7]. Juels and Wattenberg [8] presented the first fuzzy commitment sachems by combining wellknown techniques from the areas of errorcorrecting codes and cryptography, in which a cryptographic key is de-committed using biometric data. Though this scheme worked well, but it has two major shortcomings; first, it does not allow modifications of the key. Second, the security proof holds if the key is uniformly distributed. Juels and sudan fixed these drawbacks by
used directly as password or cryptography secret
proposing a fuzzy vault scheme [9]. The main
[2].
drawback of this scheme is that if it’s used for When biometric data are used in an application
it has to be stored in a database. This data might be used across a network for matching against
different application with different vault each times it reveals fingerprint minutiae. To overcome the drawback of the fuzzy
reference database. Due to this basic step, the
commitment and the fuzzy vault Yevgeniy et al
biometric system gets exposed to a new security
[10] defined a fuzzy extractor, which will be used
risk such as: constructing false biometrics from
in this paper to convert the fingerprint local feature
into a public key which can be used by any security
distributed string. The seed for the Fuzzy Extractor
application as a cryptographic key. In
along with the Sketch is stored publicly. When a
preprocessing step, skeletonization and
query needs to be matched to the input, the Fuzzy
normalization of binary images were carried out.
Extractor uses the public Sketch of the input along
Ideally the width of the skeleton is one pixel but
with the query to exactly reconstruct the input. The
this is not always true especially at the connection
Fuzzy Extractor’s reconstruction procedure is
point it produced unwanted superior minutiae
designed in such a way that if the query is within a
points, therefore, they should be removed.
specified distance from the input, the reconstruction succeeds. The reconstructed input is
2. Fuzzy Extractor Yevgeniy et al [10] defined a Fuzzy Extractor,
then mapped to the same string using the same seed stored along with the Sketch. The public storage of the Sketch and the seed
which extracts uniform string R from its input W’
do not substantially compromise the security of the input as they cannot be used to recover the input without a query which is ‘close’ to the input.
the string R can be reproduced as long as the input remains reasonably close to the original. The fuzzy
3. Propose Preprocessing Fingerprint is the oldest biometric-based identification system [11]. Because of its
extractor output a non secret string ρ to assist in
uniqueness, accessibility, reliability, this biometric system is being used widely in numerous applications, for example forensic test, security, personal identification. When a finger is pressed
reproducing R from W’.
against a smooth surface the fingerprint is produced. The most visual characteristics of a fingerprint are a pattern of ridges and valleys, they run in parallel; sometimes they gets terminated and
They introduced two primitives called a Secure
sometimes they are bifurcated.
Sketch which allows recovery of a shared secret
There are two type of feature for fingerprint
given a close approximation thereof, and a fuzzy
recognition: global feature and local feature. Global
Extractor which extracts a uniformly distributed
feature form special pattern of ridge and furrows,
string R from this shared secret in an error-tolerant
which are called singularities or singular point, and
manner.
the important points are the core and the delta. The
During the enrolment of an error prone, none uniformly distributed input; the Secure Sketch generates some public information related to the input called a ‘Sketch’ which by itself cannot be used to recover the input. The Fuzzy Extractor is used to map the non uniform input to a uniformly
core is defined as the most point on the inner most ridges; and a delta is defined as the point where three flows meet [12]. There have been several approaches for singular point detection in the literature [13; 14; 15; 16; 17; 12; 18; 19; 20; 21] and most of them operate on the fingerprint orientation field [13; 14; 15; 17; 12; 18; 19]. The
Poincare index (PI) method [12; 22; 18; 23] is one
connectivity of the ridge structures while forming a
of commonly used methods to detect singular
skeletonized version of the binary image. This
point. Minutiae are important at the local feature; it
skeleton image is then used in the extraction of
carries information about the individuality of the
subsequent feature.
fingerprint. Minutiae, in fingerprint context, are the various ridge discontinuities of a fingerprint. The ANSI [24] classified minutiae to four classes:
3.2 Feature Filtering A post-processing stage is important to remove
termination, bifurcations, trifurcations, and
the spurious minutiae introduced by the previous
undetermined. Consistent extraction of these
process or caused by dirt, excess or lake of ink, or
features is essential for fingerprint recognition.
from the fingerprint scanner. Figure (2) illustrate
There are many approaches in the literature about
the false minutiae and shows how the ridge is
locating the local feature [25; 26; 27; 2; 28; 29; 30;
connected. The facing endpoint in (a, b) are
31].
connected, bifurcation facing with endpoint is
In this paper Normalization, Thinning, Feature Filtering, and Feature Extraction are used for extracting the fingerprint feature as vectors which are used as input for the fuzzy extractor.
3.1 Normalization Normalization is used to remove the effects of
removed (c), bifurcation facing other bifurcation is removed (d), spurs are removed (e), bridges are removed (f), triangles are removed (g), and ladder structure is removed. Xiao and Raafat [33] introduced an algorithm to describe and remove the spurious minutiae: In order to describe the minutia structures, some
the sensor noise and finger pressure difference. The
statistical attribute about the fingerprint ridge are
ridge and valley structure will not be affected by
calculated as follows:
this process. It determines the new intensity value of each pixel in the image by making use of following mathematical relation:
The coordinates (x,y) of the fingerprint image are calculated from top left (0, 0) to bottom right (255,255) with the first horizontal line of scanning as the X-axis direction. The ridge length Lp1p2 equals the tracing steps
Where I'[x,y] is a 2D finger print and m and v are the image mean and variance and m0 and v0 are the desired mean and variance obtained after the
from point P1 to point P2 along a thinned fingerprint ridge. The ridge direction Dp can be obtained from
normalization [11]. The proposed approach also
adjacent ridge points of point P Tracing t steps
makes use of thinning process followed by the
(λ
