Simple Hashing for Core Point and Minutiae Based

Fingerprint Identification
S.H. Utami
#1
, Gede Indrawan
#2
, Benhard Sitohang
#3

#
School of Electrical Engineering and Informatics, Institute Technology of Bandung,
10
th
Ganeca Street, Bandung, Indonesia.
1
[email protected]
2
[email protected]
3
[email protected]

Abstract There are many existing direct access method for
fingerprint identification. A simple direct accessing method
design to observe reducing number of matching that can be
performed. With this simple hashing, we expect not only number
of matching will be reduced, but also reduce complexity of hash
function. This method design based minutiae, as local level
component, and core point, which is global level component as
reference point. The result is, there will always be a collision in
this method, but as long as stored fingerprint distribution good,
performance, speed and accuracy, will be good either.

Keywords Fingerprint identification, direct access, simple
hashing, reduce number of matching, reduce number of
complexity, minutiae, core point, distribution, performance.
I. INTRODUCTION
Fingerprint is one of biometric elements that are widely
used. Fingerprint has been used for systems which provide
community service, e.g. e-KTP (Indonesia), IAFIS (FBI,
America), CCRTIS (Canada), etc. That systemwould need
very large storage to store fingerprints data. In this case,
sequential based identification method requires system to
perform matching for every single fingerprint in database.
Fingerprint matching is a complex mechanism that involves
a lot of components and uses some complex procedure in
computation.Sequential method which complexity is O(n) is
not reliable anymore for identification in large scale
fingerprint storage.In this case, there is a need to design a
direct access method to reduce the number of matching
performed.In this research, we design a simple way for
fingerprint mapping.
There are some steps in fingerprint recognition, which
include image processing, file organization, and matching.
Focus of this research is file organization, which included
fingerprint distribution and some factor that affect fingerprint
distribution. Fingerprint matching only used for support
designed access method.
In the second section, there is some related work to this
research, included fingerprint components, fingerprint
matching, and previous fingerprint identification method.
Design of hashing method is elaborated in third section.
Testing steps, testing results, and results analysis is written in
fourth section and fifth section. In the last section, there is the
conclusion of this research.
II. RELATED WORK
Some correlated theory is needed to design hashing for
fingerprint identification, included direct access method,
fingerprint components, and existing direct access methods.
Direct Access Method
Direct access method is an alternative for inefficient
problem that perform in sequential method. In identification,
sequential search all object to get some information which
related to input, which is not efficient for system [1]. Direct
access has some clue for location of retrieved information
which means there is no need to search every single object in
storage anymore. Direct access complexity in identification is
up to O(x)[2].
There is some type of direct access method, but hashing,
indexing and direct addressing are the popular one.Direct
access can be grouped into unique value and the type which
can have similar value. Indexing and direct addressing are the
type that can be grouped into unique value. This type
hasunique value for every single input. But, in hashing two in
put or more can mapping into similar value (collision). This
design tolerance the collision of data which make hashing
decided to be designed access method. Hashing store data into
some bucket based on hash value which calculate using a
function called hash function.
Fingerprint Components
Proposes hashing is using some component which occurred
in every fingerprint. There are three levels of component in
fingerprint, which is global level, local level, and very local
level component[3]. Global level is the component that
represents the whole fingerprint image. Singular area and core
point is the two components in fingerprint. In a fingerprint,
more than one singular area may be detected, but there is only
(maximum)one core point detected, as shown in figure 1.
There is only one component that occurs in local level
component, which called as minutiae. Definition of minutiae
is an ending of a ridge pattern. Based on Galtons
classification, there are six types of minutiae, but ridge ending
and bifurcation are the more often minutiae that using for
fingerprint recognition.



JurnalSarjanaInstitutTeknologiBandungbidangTeknikElektrodanInformatika
Vol.1,No.3Oktober2012
V
iden
fact
com
iden
only
Fr
that
com
com
sign
beca
finge
minu
hash
Exis
Acce
T
for
inde
That
valu
direc
resu
the e
In
prop
Inde
isa l
mak
iden
In
acce
com
wan
we
appr
thing
A
rece
proc
fig.
Very local
ntifiedfrom ima
makes iden
mponent needs
ntification base
y developed for
rom the fact in
we will no
mponent, so the
mponent. But,
ificant access
ause left loop
erprint popula
utiae as the m
hing.
sting Fingerpr
ess
here is some re
fingerprint ide
exing method p
t method is u
ue for fingerp
ction, minutiae
ult shown that,
effect of reduci
n the other h
posed by J ia
exing based on
lot of informa
kesaccuracy ve
ntification.
III. HASHING F
F
n this previou
ess methods. T
mplicated way f
t to apply a si
can observe n
roach. Fingerp
g to observe fo
As shown in fig
ive informatio
cess. After t
1 Global Compon
fingerprint c
age with qual
ntification me
large storage
ed on fingerpr
r experimental
n the previous
ot design has
candidates on
hashing base
s time differ
p, right loop,
ation each [4].
main compone
rint Identifica
esearch which
entification. O
proposed by B
using three mi
print. Feature
e type, and ma
proposed meth
ing number of
hand, there i
anjiangFeng
minutiae and r
ation that cont
ery well, but th
FOR CORE POIN
FINGERPRINT I
us section, exp
There is a lo
for fingerprint
imple approach
numbers of m
rint distributio
or predicting re
gure 2, for sto
on about used
that, hash va
nent of Fingerprint
component c
lity minimum
ethod which
for fingerprin
rint componen
needs.
s paragraph, w
shing based
nly global level
ed global lev
rence to sequ
, and whorl
.In conclusion
ents that is us
ation Method
design a direct
One of them i
BirBhanu and X
inutiae for ca
used in ind
aximum side fr
hod has good
matching cand
is another ind
and AnniCai
ridges around m
tained in this
his effect bad
NT AND MINUT
IDENTIFICATION
plained about
ot of method
mapping. In th
h for fingerpri
matching reduc
on will be the
eduction numbe
oring fingerprin
d component f
alue calculati

t
can only b
1000 dpi.Tha
is used thi
nt. That make
nts in this leve
we can conclud
on very loca
l and local leve
el do not ha
uential metho
have 30% o
n, we will use
sed in designe
Using Direc
t access metho
is triplet base
Xuejun Tan [5
alculating inde
dexing isangle
rom triplet. Th
performance a
didate.
dexing metho
i[6]. Propose
minutiae. Ther
method whic
access time fo
TIAE BASED
N
existing direc
which is use
his research, w
int mapping, s
ction using thi
most importan
er.
nt, system mus
from extractio
ing to decid
be
at
is
es
el
de
al
el
as
od
of
ed
ed
ct
od
ed
].
ex
e,
he
as
od
ed
re
ch
or
ct
ed
we
so
is
nt
st
on
de
destination
selected c
stored into
Figure
designed f
hashing m
calculate
destination
matching
fingerprin
more deta
selected
extracted.
about fing

In sec
componen
tolerance
Core poin
make desi
core poin
function.
n bucket for
omponents, bu
o destination bu
fig. 2 Fi
3 show that, t
fingerprint iden
mechanism whi
hash value
n bucket exi
mechanism,
nt in destinatio
ail information
components,
If fingerprint
gerprint and fin
fig. 3Finge
cond section
nt used for des
translation an
nt decided to be
ign hashing ba
nt. Figure 4
storage. Hash
ut every single
ucket.
ingerprint Storing
there are two
ntification met
ich using only
for decide
ists, second
is used for
on bucket. Mat
n, so this mec
but also ev
found in the b
ngerprint owne
erprint Identificatio
written that
ign hash funct
nd rotation a r
e reference poi
ased on two co
showillustrati
value calcula
components ex
Method
different mech
thod. First mec
selected comp
destination b
mechanism,
searching eve
tching mechan
chanism not o
ery single c
bucket, then in
r is shown.
on Method
the main f
tion is minutiae
reference poin
int for fingerpr
omponents, min
ion of design

ated using
xtracted is

hanism in
chanism is
ponents to
bucket. If
which is
ery single
nism need
only using
omponent
nformation

fingerprint
e. But, for
nt needed.
rint, which
nutiae and
ned hash


fig. 4Ilustration for Simple Hash Function
For calculate hash function core point and minutiae position is
needed. Area with radius r from core point is marked as
function area f. Then, the distance from core point c to every
single minutiae m
i
in area f accumulated. Designed hash
function is very simple if compared to indexing methodin the
second section. Hash function shown using equation 1.

b: = J

m
=1
, untuksctiopJ

r (eq. 1)

Matching mechanism use complete minutiae extraction
result. Matching design using geometric approach, especially
minutiae distance. In matching two fingerprint, fingerprint
with more minutiae number become the reference F1 and the
other called F2. N random minutiae selectedfa
i
from fb. Every
fa
i
is compared with every minutiaefb
i
in F2 which is
calculating using the number of simultaneous matched
minutiae distance R.Find nearest minutiae from fa
i,
cai, and
nearest minutiae from fb
i
, cb
i
, then compare their types and
distances.If pair of fa
i
and ca
i
match with pair of fb
i
and cb
i
,
than find nearest minutiae from ca
i
and cb
i
for next
comparison. Match rate R is add for every matched pair of
minutiae which comparison start form fa
i
and fb
i
. If match rate
reach the minimum rate, R
min
, than F1 and F2 match. Here is
the pseudocode for matching mechanism.

Mat ch( F1, F2)
i f ( F1. mi nut i aenumber >=F2. mi nut i aenumber )
{
Ref er ence = F1;
Pai r = F2
}
el se
{
Ref er ence = F2;
Pai r = F1;
}

/ / Mat chi ng
f a = Ref er ence[ 0] ;
Whi l e ( not mat ch and not end of Ref er ence and
r at e<mi nRat e)

{
Whi l e ( not mat ch and not end of Pai r and
r at e<mi nRat e)
{
f b = Pai r [ 0]
ca = f a; cb = f b;
i f ( mat cht ype( r a, r b) and mat cht ype( ca, cb)
and mat chdi st ance( r a, r b, ca, cb) )
{
mat ch = t r ue;
++r at e;
r epeat
{
r a = ca; r b= cb;
ca = near est mi nut i ae( r a) ;
cb = near est mi nut i ae( r b) ;
i f ( mat cht ype( r a, r b) and
mat cht ype( ca, cb) and mat chdi st ance( r a, r b, ca, cb) )
{
++r at e;
mat ch=t r ue;
}
el se
{
mat ch = f al se;
r at e = 0;
}
}
unt i l ( not mat ch and f i ni shed( f b) and
r at e>=mi nRat e)
}
++f b;
}
f a = f a + addnumber ;
}


IV. TESTING
Testing use fingerprint from FVC 2000 db1_a[7] and FVC
2002 db3_a[8] which resolution is 300x300 pixel. Minutiae
extract (only bifurcation) from DatabaseAnalyzermodul in
SourceAFIS[9].Core point extract from ortcorepoint[10]. For
feature transformation, which used for identification data test,
we make generator for transformation of an extraction result.
Designed method is tested using software which built for
testing purpose, called FHIS (Fingerprint Hashing for
Identification Simulator). Here is some observed object in
testing:
a. Distribution of stored fingerprint and some factor that affect
fingerprint distribution, included radius of hash function
area and number of feature in hash value.
b. Method performance, included accuracy, which means
tolerance factors, and number of matching process.
Performance will be observed during identification process.









V. RESULT AND ANALYSIS
A. Distribution of Stored Fingerprint
In this subsection, analysed effect from some factors, radius
value, number of variable used as hash value, and
fingerprints characteristic, to the distribution of stored
fingerprint.
Table 1 shows the radius effect to fingerprint distribution.
In choosing radius value, there is a need to take most
minimum value possible because that will increase the
tolerance for fingerprint modification (transformation) in
identification input. For the first try, 10 pixel isused as radius
value for hash function area. But there is a problem, a lot of
fingerprints are mapped into 0 bucket, which means many
fingerprints dont have any minutiae in hash value area. After
the second try with radius 20 pixel do not give a good enough
data distribution, an analysis is done for the test set. Data set
analysis result shown that average of minimum distance
minutiae to core point in fingerprint is from 30s to 40s pixel.
To minimize collision in 0 bucket, then, radius 50 pixel is
used and the result for fingerprint distribution is good enough.
There are only 26.4% fingerprints that are saved into existing
bucket.
TABLE 1
RADIUS EFFECT TO FINGERPRINT DISTRIBUTION
Comparison Radius of Hash Function Area
10 Pixel 20 Pixel 50 Pixel (After
fingerprint
position
analysing)
Percentage hash
function area
from whole
fingerprint image
0.17% 0.7% 4.36%
Fingerprint
distribution
There
aremanyfingerprints
in bucket with hash
value 0.
Less collision
in 0 bucket,
but the
numbers are
still many.
Distribution is
good. Only 4
maximum
fingerprint in a
bucket (2%
from data test),
which is in 0
bucket and -1
bucket (no core
point bucket)

Hypothetically, more feature use as hash value, more
buckets will be formed. In testing, we modified the hash value
with added the number of minutiae in hash function area as
the other feature. After do experiments for radius 10 pixel, 20
pixel, and 50 pixel, we conclude that with added number of
selected minutiae as hash value feature, the number of
fingerprints in 0 bucket cannot be reduced. But, with added
feature, distribution in other bucket is better. The point is a
long as distribution using less features is good, there is no
need to add feature, because that will make hashing
computation more complicated.



TABLE 2
NUMBER OF FEATURE USED IN HASH FUNCTION EFFECT TO FINGERPRINT
DISTRIBUTION
Comparison Feature in Hash Function
Distance Distance and
Number of Selected
Minutiae
Total fingerprints are
saved into existing
bucket.
26.4% 10.15%
Maximum number of
fingerprints in a bucket
4(2%) in 0 bucket and
-1 bucket
4(2%) in 0 bucket and
-1 bucket

In data set, distance from nearest minutiae to core point
further than the distance between minutiae (bifurcation). So, if
one of bifurcation is used as reference point, there is a chance
that area hash function is smaller.
B. Performance of Simple Hashing
From observing identification process, the result is number
of matching in identification process depends on
storedfingerprint distribution.For example, identification in
hashingusing radius 50 pixel, which is much better than
hashing using radius 10 pixel.
For better accuracy, after get the hash value from input
fingerprint, there is a need to check +n bucket around the
exact bucket. Based on testing result, n value is 5 pixel which
around 1.8% from hash value range of data test. This make
number of matching increase into 3 to 5 times bigger.
Tolerance n is given from estimation of average minutiae
selected in calculation of hash value, because error occurs
from the rounding off in distance calculation.
Scaling is not tolerance in this method. But, translation and
rotation is accepted as long as hash value area is not damaged.
VI. CONCLUSIONS
Simple hashing can be used to reduce number of matching
in identification. But, some factors that affect distribution of
stored fingerprint must be estimate carefully to get the
maximum performance. In identification there is some error
tolerance that must be given to select identification candidate
for matching.
REFERENCES

[1] A. Silberschatz, H. F. Korth, and V. Govindarahu,
Database System Concept 6th Edition. New York:
McGraw-Hill, 2011.
[2] T.H. Cormen, C.E. Leiserson, and C. Stein, Introduction
to Algorithm 2nd Edition. Massachusetts: MIT Press,
2002.
[3] D Maltoni, Handbook of Fingerprint Recognition 2nd
Edition. London: Springer, 2000.
[4] C Wilson, G. Candela, and C. Watson, "Neural Network
Fingerprint Identification," Journal Artificial Neural
Network, vol. 1, no.2, pp. 203-208, 1993.
[5] B. Bhanu and X. Tan, A Triplet Based Approach for
Indexing of Fingerprint Database for Identification.

Riverside: University of California, 2001.
[6] J . Feng and A. Cai, Fingerprint Indexing Using Ridge
Invariant. Beijing: Beijing University of Posts and
Telecomunications, 2006.
[7] D. Maio, D. Maltoni, and L. Capelli. (2000, August)
FVC2000. [Online]. http://bias.csr.unibo.it/fvc2000
[8] D. Maio, D. Maltoni, and L. Capelli. (2002, April)
FVC2002. [Online]. http://bias.csr.unibo.it/fvc2002
[9] R. Vazan. (2009, December) Sourceforge. [Online].
http://sourceforge.net/projects/sourceafis
[10] L. Rosa. (2012, April) Advanced Source Code. [Online].
http://wwwadvancedsourcecode.com/ortcorepoint.pdf