Fingerprint _prem

354 IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 17, NO. 3, MARCH 2008

Local Features for Enhancement and Minutiae
Extraction in Fingerprints
Hartwig Fronthaler, Klaus Kollreider, and Josef Bigun, Senior Member, IEEE

Abstract—Accurate fingerprint recognition presupposes robust locally constant direction [2]. In realistic scenarios though,
feature extraction which is often hampered by noisy input data. the quality of a fingerprint image may suffer from various
We suggest common techniques for both enhancement and minu- impairments, caused by 1) scares and cuts, 2) moist or dry skin,
tiae extraction, employing symmetry features. For enhancement,
a Laplacian-like image pyramid is used to decompose the original 3) sensor noise/blur, 4) wrong handling of the sensor (e.g.,
fingerprint into sub-bands corresponding to different spatial too low/high contact pressure), 5) generally weak ridge-valley
scales. In a further step, contextual smoothing is performed on pattern of the given fingerprint, etc. While not acting against
these pyramid levels, where the corresponding filtering directions low-quality fingerprints, methods to automatically assess the
stem from the frequency-adapted structure tensor (linear sym- quality of a given impression, such as [3]–[5], are useful
metry features). For minutiae extraction, parabolic symmetry is
added to the local fingerprint model which allows to accurately and complementary to this study. The task of fingerprint en-
detect the position and direction of a minutia simultaneously. Our hancement is to counteract the aforesaid quality impairments
experiments support the view that using the suggested parabolic and to reconstruct the actual fingerprint pattern as true to the
symmetry features, the extraction of which does not require original as possible. The latter part is especially noteworthy.
explicit thinning or other morphological operations, constitute Furthermore, unrecoverable areas should be labeled as such,
a robust alternative to conventional minutiae extraction. All
necessary image processing is done in the spatial domain using since the enhancement at too noisy parts may generate spurious
1-D filters only, avoiding block artifacts that reduce the biometric features. There are several published studies on fingerprint
information. We present comparisons to other studies on enhance- image enhancement. Hong et al. [6] proposed an algorithm
ment in matching tasks employing the open source matcher from using Gabor bandpass filters tuned to the corresponding ridge
NIST, FIS2. Furthermore, we compare the proposed minutiae frequency and orientation to remove undesired noise while
extraction method with the corresponding method from the NIST
package, mindtct. A top five commercial matcher from FVC2006 preserving the true ridge-valley structures. All operations are
is used in enhancement quantification as well. The matching error performed in the spatial domain, whereas the contextual fil-
is lowered significantly when plugging in the suggested methods. tering in [7] and [8] is done in the Fourier domain. Either way,
The FVC2004 fingerprint database, notable for its exceptionally blockwise processing is used to obtain the enhancement result
low-quality fingerprints, is used for all experiments. causing restoration discontinuities at the block boundaries.
Index Terms—Differential scale space, fidelity, fingerprint These methods are likely successful in extremely bad-quality
restoration, image enhancement, image pyramid, linear symmetry, regions, but also rather rigid under easy conditions, due to the
minutiae extraction, orientation tensor, parabolic symmetry, sym-
radical processing. In [9] and [10], the fingerprint’s blockwise
metry features.
Fourier transform is multiplied by its power spectrum raised
to a power , thus magnifying the dominant orientation. A
I. INTRODUCTION blockwise Fourier transform is also employed by Chikkerur
et al. [8], [11], followed by contextual filtering using raised
cosines. In a related study [12], a standard discrete scale space
OR THE processing of fingerprint images, two stages
F are of pivotal importance for the success of biometric
recognition: image enhancement and feature/minutiae extrac-
has been used to contextually process fingerprints. We employ
a multigrid representation of a discrete differential scale space
following a novel enhancement strategy. For a more detailed
tion. A topical review of fingerprint processing technologies review of fingerprint enhancement schemes, we refer to [2].
can be found in [1]. Applied immediately after sensing but In this study, we propose the use of an image-scale pyramid
before feature extraction, an optional image enhancement and directional filtering in the spatial domain for fingerprint
can be performed to facilitate the feature extraction and the image enhancement to improve the matching performance as
subsequent processing by “de-noising” the signal. In an ideal well as the computational efficiency. Image pyramids or mul-
fingerprint image, ridges and valleys alternate and flow in a tiresolution processing is especially known from image com-
pression and medical image processing [13], [14], but their rel-
Manuscript received May 8, 2007; revised November 28, 2007. The associate evance to fingerprint image enhancement has not been quanti-
editor coordinating the review of this manuscript and approving it for publica-
tion was Dr. Peyman Milanfar.
fied before. The Laplacian pyramid [15], [16] is equivalent to
The authors are with Halmstad University, SE-30118, Halmstad, bandpass filtering in the spatial domain. Here, we decompose
Sweden (e-mail: hartwig.fronthaler@ide.hh.se; klaus.kollreider@ide.hh.se; fingerprint images in a similar manner, since we expect all the
josef.bigun@ide.hh.se). relevant information to be concentrated within a few frequency
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. bands. Furthermore, we propose Gaussian directional filtering
Digital Object Identifier 10.1109/TIP.2007.916155 to enhance the ridge-valley pattern of a fingerprint image using
1057-7149/$25.00 © 2008 IEEE

FRONTHALER et al.: LOCAL FEATURES FOR ENHANCEMENT AND MINUTIAE EXTRACTION 355

computationally cheap 1-D filtering on higher pyramid levels where , and and denote the derivatives of the
(lower resolution) only. The filtering directions are recovered image in and direction, respectively. The rightmost part
from the orientations of the structure tensor [17], [18] at the cor- of (1) indicates the actual implementation with a Gaussian bell
responding pyramid level. Linear symmetry features are thereby , and “ ” denotes a 2-D convo-
used to extract the local ridge-valley orientation (angle and reli- lution. The operations and are realized by means of
ability). In contrast to other studies, e.g., [6] and [11], no block- convolutions via and (1-D) with C
wise processing is performed, avoiding block boundary arti- being the nonessential constant . The calculation of the
facts. tensor implies complex filtering but the synthesis of the com-
Furthermore, we focus on minutia points, which are disconti- plex happens afterwards at pixel level. This is also true for
nuities in the ridge flow. The most prominent minutiae are ridge the symmetry features to be explained next. A symmetry in the
bifurcation and termination ( ending), which refer to points image and the associated filter detecting it can be modeled by
where a ridge divides into two and a ridge ends, respectively. In , where represents its symmetry order [27]–[29].
what follows, we benchmark a recent method to detect the minu- Parameter can be any integer, including negative ones. It was
tiae’s position and direction [19]. Existing minutiae extraction shown that to detect a symmetry of type in an image, one
approaches comprise so-called “direct grayscale” and “binariza- can convolve , described by (1), with the complex filter given
tion-based” methods. In “binarization-based” methods [20], the by the following polynomial:
fingerprint image is binarized and morphologically analyzed for
minutiae. In “direct grayscale” methods, mainly the image gra- (2)
dient and local grayscale neighborhood are used to locate minu-
tiae, either by tracking ridges [21], or by classifying directional Even symmetries of the type can be detected in the
filter responses [22]. The direction of a minutia is mostly inher- same way. The corresponding filter is then given by the com-
ited from the associated ridge. The presented method is a “direct plex conjugate of the -filter in (2). The first type of sym-
grayscale” approach, as it operates exclusively on the local di- metry used in this study is linear symmetry, which occurs at
rection field of a fingerprint image. Note that with partial finger- points of coherent ridge flow. It naturally encodes the local ori-
prints lacking singular points [23] at hand, only local features entation and coherence of the ridge-valley structure as a com-
can be used for alignment and matching purposes. We extend plex number. The discrimination power of this information has
the features used to enhance the image by higher orders sensi- been shown in fingerprint pattern matching [30]. In the case of
tive to parabolic symmetry which enable us to detect the minutia linear symmetry, which can also be described as symmetry of
points’ position and direction. The targeted minutia types are order the polynomial in (2) reduces to a Gaussian filter
ridge bifurcation and termination. Linear and parabolic sym- , where . Applying this filter to the orientation
metry are the only features used throughout our study. tensor corresponds to complex summation/averaging of
We corroborate our techniques by experiments on a state-of- the linear symmetry tensor in the local neighborhood of each
the-art database for fingerprint recognition, the FVC2004 data- point. To obtain a reliable measure for the linear symmetry, we
base [24]. We use the matching module of the NIST FIS2 soft- first calculate the second order complex moments
ware [25], [26] to compare our features to the binarization/mor- and , as suggested by [17] and [31]. The measure
phology based features of the software, in an attempt not to for linear symmetry used in this study is denoted in (3)
favor our features by using a commonly referred matcher. Fur-
thermore, we compare two different fingerprint enhancement (3)
methods with our method qualitatively (by human vision) and
objectively (via matching performance experiments). We also
Moment acts as an upper boundary for the linear sym-
test the proposed enhancement in conjunction with a top per-
metry certainty, and by dividing through unreliable ori-
forming matcher of the FVC2006.
entations are attenuated, whereas the strong ones are promoted.
It is worth noting that linear symmetry detection is also useful
II. LOCAL FEATURES for locating minutiae in an inverse fashion, as “lack of linear
We use two types of symmetries to model and extract the local symmetry” occurs at minutia points [32] (compare Figs. 4 or 6).
structure in a fingerprint, which are parabolic and linear sym- Next, we are interested in parabolic symmetry of order
metry. Both symmetries can be estimated by separable filtering , since the parabolic pattern is most similar to a minutia point
of the orientation tensor. For a more detailed review of sym- in a fingerprint. Equation (4) describes how complex filtering is
metry filters, i.e., symmetry derivatives of Gaussians, we refer applied to detect this type of symmetry, where represents the
to [18] and [27]. The starting point for the extraction of our local 2-D scalar product (as before) and denotes the Gaussian
features is the linear symmetry tensor1 image, described by the window
complex expression
(4)

The two grayscale images in Fig. 1 show minutia points of
(1) type ridge bifurcation (top) and ridge ending (bottom), with their
1It is also known as the structure tensor or the inertia tensor which can be directions indicated. Next to them, the corresponding complex
fully represented by (1). filter responses of are displayed. The latter can be described


TABLE I
PYRAMID BUILDING PROCESS

A. Pyramid Decomposition (PD)
A pyramid decomposition requires resizing (scaling, or geo-
metric transformation). To create our Gaussian and Laplacian-
like pyramids, we define the reduce and expand
operations, which decrease and increase an image in size by
Fig. 1. Parabolic symmetry filter responses PS for (top) ridge bifurcation and the factor , respectively. During reduce, the image is initially
(bottom) ridge ending. low-pass filtered to prevent aliasing using a Gaussian kernel.2
The latter’s standard deviation depends on the resizing factor,
which here follows the lower bound approximation of the cor-
responding ideal low-pass filter, [18]. We
initially reduce the original fingerprint image by a factor of
in order to exclude the highest frequencies. In a fur-
ther step, we reduce the image size by a factor for
three times. This is also outlined on the upper left hand side of
Table I. To create images containing only band limited signals
of the original image, we expand the three images by factor
and subtract each of them from the next lower level, yielding
Fig. 2. Data flow: (rectangles) images and (connection labels) applied pro-
cessing. The abbreviations will become clear within this section.
. The latter contain the adequately high, medium and low
frequencies (ridge-valley pattern) of the original fingerprint. It
is worth noting, that only the Laplacian-like pyramid levels
are used subsequently in this study. In a further step, the contrast
as . The value is a certainty measure and
of the single band images is enhanced, following
the argument represents the geometric orientation of the sym-
where , to depreciate small vectors
metric pattern in symmetry order . An important property
of in comparison with those of large magnitude. Note, that
of parabolic symmetry filtering for minutiae detection is that the
contains no DC content, i.e., ridge pixels have negative values
minutia direction is retrieved at the same time (see Fig. 1) as its
whereas valley pixels are positive (ideally). In Fig. 3, we show
position, and that this minutia direction estimation is indepen-
example fingerprints of different quality besides their contrast
dent of the ridge direction estimation.
enhanced pyramid levels . For easier comprehension, the
Symmetry features of these and higher orders have been
latter are used for an initial reconstruction step, which is dis-
useful in a number of tasks in image processing, for example,
played rightmost. This reconstruction is crude so far and rep-
singular point extraction in fingerprints [33], image quality
resents only an isotropic (nondirectional) enhancement, which
estimation [3], texture analysis and optical flow estimation
is implemented as listed at the bottom of . It is already visible
[34], recognition of crash test cross trackers [27], etc.
that the portions of the fingerprint image that have been retained
and contrast improved contain significant recognition informa-
tion, whereas others containing high-frequency isotropic noise
III. FINGERPRINT IMAGE ENHANCEMENT
are attenuated.
It has been shown previously that the quality of a fingerprint We cover approximately half the bandwidth of the original
image directly affects the performance of a given recognition image through the bandpass images in total, e.g., by setting
system [3], [24], [35]. In lack of higher quality images, e.g., and to 1.5. This choice depends on the resolution of the fin-
from a crime scene, or due to worn fingerprints, a carefully un- gerprint and normally can be approximated off-line, either ex-
dertaken high-fidelity enhancement is frequently the only option perimentally or by using information on the used sensor. In the
that remains to enable identity establishment via fingerprints. images studied here the ridge frequency in a fingerprint image
The steps involved in the proposed strategy are arranged as il- 2The 2-D Gaussians are the only functions that are Cartesian separable and
lustrated in Fig. 2, each of them to be detailed as follows. yet being fully rotational symmetric, avoiding orientation bias [18].


Fig. 3. Pyramid decomposition process of the (top row) high-quality impres- Fig. 5. Directional filtering process of the example fingerprint images featuring
sion 104 6 and a (bottom row) low-quality example 1 1 of the FVC2000-2 high (top row) and a low quality quality (bottom row); the column images rep-
dataset; the column images represent 1) the original fingerprint, 2–3) its band- resent 1) the original fingerprint, 2–3) directionally filtered pyramid levels l
pass like decomposition via l after contrast enhancement, and 4) the “so-far” and iv) the final fingerprint after reconstruction and contrast enhancement.
reconstructed fingerprint.

meaningful because contains the most localized orienta-
tion (information also at minutia-level), but is also most sus-
ceptible to noise. In Fig. 4, for the two example finger-
prints from 3 are displayed using a HSV model, where the mag-
nitudes modulate value V and the arguments (local orientation)
steer hue H. Note the effect of low quality on , especially
in the second row. However, the higher pyramid levels ,
which contain a coarser description of a fingerprint’s orienta-
tion, are still intact. When compared to a low-pass pyramid
(e.g., Gaussian pyramid), the estimated orientations via band-
pass pyramids (e.g., Laplacian pyramid) were found noticeably
more robust, in this context.

C. Directional Filtering (DF)
To enhance the SNR (signal-to-noise ratio), i.e., to remove
sweat pores, scars, etc., we apply directional averaging to all
levels independently, described next. The local filtering di-
rection within is given by , thus it follows the
Fig. 4. HSV representation of linear symmetry filter responses LS
ridges/valleys of the fingerprint. At every point, the neighboring
(columns 1–3) for the two example fingerprints of the FVC2000-2 dataset, pixels along a line having the same local direction are averaged
where the magnitude steers V and the argument controls H. with a (1-D) Gaussian, yielding the new value. The possible
number of different averaging directions is quantized (here 20).
Additionally, we exploit the magnitudes of the complex pixels
is 60 cycles per image width/height [6]. This translates to of . First, pixels where are assigned to the back-
an image dimension of approximately 100–400 pixels for the ground, i.e., they are set to 0 (effectively amounting to a seg-
method to be most effective. mentation of the fingerprint from the background or the heavily
noisy regions). Second, only if when measured
B. Orientation Estimation (OE) on a small annulus centered at the current pixel, a reasonable
The ridge-valley orientations for each of are estimated quality (presence of ridge-valley pattern at level ) is ensured
using (3). We calculate for level . Being and the above filtering is done. Otherwise, the pixel is again
a complex valued image, encodes the local orientation in set to 0. Generally, should be chosen higher than , advis-
a pixel’s argument, , and the reliability of this estimation able values are below 0.5. In this pyramidal process, the struc-
in the magnitude, . By using LS the measure becomes ture tensors become spatial frequency selective and adaptively
independent of the signal energy. Furthermore, is attenu- steer the smoothing amount and direction along the ridge-valley
ated if its orientations deviate too much from the ones of , structures. At the lowest level , fine minutiae are preserved be-
which is done by . This is cause the filtering directions are sensitive to them. At higher


j j
Fig. 6. Minutia point detection process for an example fingerprint; 1) impression 6 0 of the FVC2000-2 dataset and 2) its enhanced version, 3) LS (HSV), 4) PS ,
j j1 0j j
5) PS (1 LS ), 6) 30 strongest minutiae.

levels the rough ridge-valley flow is smoothed, and gaps are above a threshold . Thus, we can exclude spurious minutia
closed (e.g., caused by scars) because contain the global points which occur at the transition from the fingerprint to the
orientation. By use of the filtered levels only, the image is background and at impurities, comprising regions lacking fin-
reconstructed (R). A final contrast enhancement (CE) is done gerprint structure, e.g., due to low contact pressure, humidity,
subsequently. In Fig. 5, the filtered versions of for the ex- greasiness, dust, wounds, sensor deficiency, etc., within the
ample fingerprints are displayed besides the reconstructed final, fingerprint area. If this condition is fulfilled, the position and
contrast enhanced images. is most susceptible to noise, which the complex filter response are stored. One should set very
can be observed in the bottom row, where the threshold condi- close to 0 and near to 1. The final minutiae list is ordered by
tions for and are not fulfilled in many parts of the potential magnitude of PS since it represents the presence of parabolic
matching area. Note that in this case, the smoother levels symmetry. In image VI of Fig. 6, the circles indicate detected
manage to fill these holes. Singular points like core and delta minutiae when considering only the 30 highest magnitudes.
points do not need further attention during the filtering process,
due to the small size of the employed 1-D Gaussians. In Sec- V. EXAMPLES AND QUALITATIVE OBSERVATIONS
tion VI, we will quantify the benefits this fingerprint enhance- In this section, we inspect the proposed techniques for en-
ment affords along with a comparison to alternative strategies. hancement and minutiae extraction visually, including existing
methods for both tasks. We discuss the systematic qualitative
IV. FINGERPRINT MINUTIAE EXTRACTION differences briefly.
Given the enhanced fingerprint image, both LS and PS First, we show typical examples of enhancement using the
are computed via (3) and (4) to obtain features to be used in proposed method and two other published studies by Hong et al.
matching. Some more steps have to be considered to reliably [6] and Chikkerur et al. [8], respectively. The implementation
detect minutia points. First, the selectivity of the parabolic of the two latter methods are due to the authors of [8]. Fig. 7
symmetry filter responses is improved, using the simple inhi- depicts two fingerprint images representing (top row) high- and
bition scheme , [36]. Essentially, the (bottom row) low-quality from the FVC2004-1 dataset together
parabolic symmetry is attenuated if the linear symmetry is with their enhanced versions as delivered by the mentioned
high, whereas it is preserved in the opposite case. In Fig. 6, techniques. The images in the second column, corresponding to
the minutiae detection process is visualized for an example the output of [6], are obtained by the use of Gabor-filters, tuned
fingerprint. The first two images depict the initial fingerprint to the estimated spatial frequency and orientation within small
and its enhanced version, respectively. The parabolic symmetry blocks of the fingerprint. The filtering is only performed if the
displayed in image IV is inhibited with the absolute corresponding region exhibits sufficient ridge-valley structure
value of the linear symmetry shown in image III (LS). The according to an estimated signal strength. The results of [8] are
latter image also represents the local orientation. The resulting visualized in the third column of Fig. 7. Here, all calculations
sharpened magnitudes are displayed in image V. In a are performed in the frequency domain, using short time fourier
further, step all filter responses below a certain threshold transform (STFT), involving small overlapping blocks. The
are set to zero. LS provides a good measure for segmenting the ridge frequency and orientation are determined in the Fourier
fingerprint from the background in order to discard responses domain, to steer the contextual filtering by steep bandpass func-
in the marginal area of the fingerprint image. The remaining tions. The segmentation of a fingerprint is done by comparing a
filter responses within , which are concentrated to pixel “is- block’s energy to a predetermined threshold. The final column
lands,” are used for the extraction of minutia candidates. During in Fig. 7 depicts the results of the proposed method. As to be
this extraction process, we search for the highest filter response expected, our approach does not show block-artifacts because
in a small neighborhood (here 9 9, determined empirically) the data processing is not blockwise. Qualitatively, our method
throughout the fingerprint area to avoid multiple detection of produces high contrast between ridges and valleys and the
the same minutia. Finally, we demand each minutia to be fully result generally exhibits more fidelity to the original compared
surrounded by high linear symmetry, by ensuring whether the to the alternative approaches. The latter can be well observed
average linear symmetry on a ring around a minutia candidate is in the bottom row of Fig. 7, where the alternative approaches


Fig. 7. Enhancement results for the high-quality impression 99 5 (top row) and a low-quality example 1 7 (bottom row) of the FVC2004-1 dataset; the columns
correspond to 1) the original image, enhanced by 2) Hong’s, 3) Chikkerur’s, and 4) the proposed method.

seem to give in (second column) or tamper structure (third some spurious, as well as misses some minutiae in comparison
column). The images created by the method of [6] (second to our algorithm. The NIST algorithm quantizes the minutiae
column) appear more blurred while the ones produced by [8] angles in intervals, causing visible deviations from the
(third column) contain visible block-artifacts, especially near actual minutiae directions. Additionally, the minutiae tend to
minutia points. Being the basic resource for most fingerprint be imprecisely located with a random error, which is likely due
recognition techniques, including semiautomatic forensics, it to the information loss during binarization and thinning needed
is to be expected that minutia neighborhood degradation will by that method. When facing very low-quality conditions
decrease recognition performance. (second column), the enhancement can only recover parts of
To provide an initial comparison for the proposed minutiae the original ridge/valley structure with rather low contrast. The
extraction method, we extract the minutiae with the method NIST provided method is possibly not trained on such images
suggested by the (US) National Institute of Standards and and is, therefore, not able to perform reasonable minutiae
Technology (NIST, FIS2 mindtct package), [25], [26]. This extraction in this case. Apart from some spurious and missed
method takes a fingerprint image, whether pre-enhanced or minutia points, the proposed method appears to adapt better
not, and determines its minutia points fully automatically by to these quality conditions. When taking a closer look, five of
means of binarization followed by morphological analysis. The the detected minutiae are actually valid ones. The others were
output file contains position, angle, and quality of each detected introduced by enhancement and minutiae detection errors.
minutia point. We employ two fingerprint impressions from the Next, we will detail experimental results supporting the qual-
FVC2004-3 dataset for a visual comparison, which are shown in itative observations above.
the first row of Fig. 8. The first image is of good quality, whereas
the second one represents very low-quality conditions intended
VI. EXPERIMENTS
to pinpoint the limits of both fingerprint image enhancement
and minutiae extraction (worst case scenario). The images in the To quantify the capability of the suggested features and
second row of Fig. 8 depict the local minutiae extraction results enhancement method we need to test them on very low-quality
of the NIST provided method, whereas the third row contains fingerprint data, where reliable enhancement and minutiae
the corresponding result of our minutiae extraction method. extraction becomes indispensable. Therefore, we use the FVC
Both impressions were pre-enhanced by the introduced method 2004 database [24], which was collected to provide a more
and the detected minutia points are superimposed (rings) on challenging benchmark for state-of-the-art recognition systems
the enhanced fingerprint images. Minutia angles are indicated than previous fingerprint verification competitions [37]. When
by directed lines originating from the circles. In the case of collecting the fingerprint data, individuals were asked among
(first column) high quality , the method of [25] and [26] detects other things to vary the contact pressure applied to the sensor


TABLE II
EERS OF THE NIST FIS2 MATCHER ON THE ORIGINAL AND PRE-ENHANCED
FVC2004 DATABASE, FOR DIFFERENT ENHANCEMENT METHODS

TABLE III
EERS, FMR100S, FMR1000S, AND ZEROFMRS OF THE COMMERCIAL
MATCHER ON THE ORIGINAL AND PRE-ENHANCED FVC2004 DATABASE

of two fingerprints (the bozorth3 package). Needless to say,
pre-enhancement should lead to lower error rates for the finger-
print matcher, compared to when trying to match the original
fingerprints. Table II shows the equal error rates (EER) of this
setup using the full FVC2004 database (all four datasets). The
equal error rate marks a system’s operating point, at which it
erroneously recognizes genuine users and imposters with equal
probability. The results in the first row were achieved when
using no image enhancement at all. The other rows represent
equal error rates in case of all impressions being initially en-
hanced by the method of Hong (second row), Chikkerur (third
row), and by the proposed approach (last row). Interpreting
Table II, we conclude that it is not self evident that what is
Fig. 8. Minutiae extraction applied to the (left column) high-quality impres-
sion 7 1and (right column) a very low-quality example 42 8 of the FVC2004-3 perceived as an enhancement improves the recognition perfor-
dataset; The three rows show 1) the original fingerprint, minutiae’s positions mance. By contrast, there is a significant risk that it actually can
and angles detected by 2) the NIST FIS2 mindtct package versus 3) the pro- worsen (!) the recognition performance, especially when the
posed method. The fingerprint images were enhanced by our method and the
second column shall also indicate its limitations. images are noisy. The method suggested here clearly has led
to favorable recognition rates among the tested enhancements,
resulting in the lowest EERs on all four datasets. In Section VII,
and their fingers were additionally dried or moistened to en- we will discuss the underlying reasons in further detail. No
force challenging image quality conditions. The FVC 2004 parameters of our enhancement method have been specially
consists of four datasets, which were acquired using different adapted to any of the used datasets, neither were the others.
sensor types and each of them contains eight impressions of Studying the matching failures, we could conclude that the
100 fingers. Subsequently, we will refer to these datasets as fingerprint area useful for recognition sometimes happened to
DB1-4. It is worth mentioning that DB4 was created using the be very small, when enhanced by the other methods (especially
SFinGE synthetic fingerprint generator [38] whereas DB1-3 are in DB1). This is indicated by the parentheses in Table II.
populated by images representing authentic fingerprints sensed As to signal enhancement, it may seem easy to improve
by real sensors. When carrying out fingerprint verification for recognition rates of a modest matcher, or to show qualitative
a single dataset, we follow the FVC protocol involving 2800 advantages over other research prototypes. For these reasons,
genuine trials and 4950 impostor trials. we assisted fingerprint recognition systems which were al-
In our first experiment, we pre-enhanced all fingerprints ready state-of-the-art in all respects. A top five matcher of the
of DB1-4 with three different enhancement methods: The FVC 2006 benchmark, termed commercial matcher below,
proposed method, and the ones of Hong [6] and Chikkerur [8], was reevaluated on fingerprints that were pre-enhanced by our
which we already illustrated by way of examples in Section V. method. Following the FVC evaluation style, we show full ROC
We employed the complete NIST FIS2 fingerprint matcher, curves and explicitly list some operating points in this case.
consisting of a minutiae extractor (the mindtct package) and a Three such points are particulary important, besides the EER,
method that evaluates the correspondence of the minutiae sets the FMR 100, FMR 1000, and ZeroFMR. FMR stands for false


Fig. 9. ROC curves of the commercial matcher on the original and pre-enhanced (by the proposed method) datasets of FVC2004. The EER is to be read off at the
intersection point between a curve and the diagonal line.

match rate, the probability of an impostor being erroneously TABLE IV
recognized as genuine user. Furthermore, FMR 100, FMR 1000 EERS OF THE NIST MATCHER WHEN USING ITS OWN MINUTIAE VERSUS
USING MINUTIAE EXTRACTED BY THE SUGGESTED METHOD. THE FVC2004
and ZeroFMR are popular marks meaning 1 false match in 100 DATABASE IMAGES WERE PRE-ENHANCED BY THE METHOD PRESENTED IN
trials, 1 in 1000 and 0 such erroneous decisions. The lowest SECTION III IN BOTH CASES
false nonmatch rate (FNMR) that a system can achieve when
its FMR is below or equal to the mentioned amounts, is repre-
sented by these three indicators. In other words, they determine
how often a genuine user is not recognized as such, at a certain
security level, and are sometimes given more significance than by our method, to the NIST minutiae matcher (bozorth3) is not
the EER by systems used in practice. This is because these less than fair because one would assume that the latter was de-
measures are more indicative for the user annoyance to be ex- signed to cope best with its own minutiae. It is also worth noting
pected. In Table III, we have listed the recognition rates of the that there are many ways of doing a matching given two sets of
commercial matcher on the original and pre-enhanced datasets minutiae [39], and that this was the only way available to us
DB1-4. Next to these, we show the relative variation between of quantifying the proposed technique for minutiae extraction.
each pair of rates (columns Original serve as reference). The In Table IV, second row, the corresponding EERs, employing
enhancement method was the one proposed, with the difference the proposed minutiae extraction method are given, whereas we
that the downsize factors had been adjusted once to restate the error rates involving the NIST feature extraction in
match the fingerprints (dimensions) more accurately. Recalling the first row (the top row of Table IV and the bottom row of
Table II, one can observe that the commercial matcher achieves Table II are naturally the same). We can observe that the pro-
remarkably lower EERs than the one of NIST. This behavior posed scheme for minutiae extraction leads to improved recog-
is valid also at other operating points. We omitted an explicit nition on all four datasets. The improvements are substantial and
list of those since the purpose here is not to compare the two meaningful because neither of the methods (for enhancement
matchers, though. Studying Table III reveals that the proposed and minutiae extraction) were specially adapted to any of the
enhancement leads to decreased error rates of the commercial tested datasets or the NIST matching module.
matcher at all listed operating points and on each of the tested
datasets. The best results were observed on dataset DB2 (high- VII. DISCUSSION
lighted column), for which the error rates dropped by over 40%, Generally, it is difficult to quantify the advantages of local
measured in percentage point drop-off. In Fig. 9, we further features used for recognition, other than by the success of known
show the corresponding receiver operator characteristics (ROC) procedures and contexts employing them. Using such proce-
curves, which extend the analysis to all possible operating dures on publicly available databases allows testing the methods
points. The axes of such a plot represent the false match and in numerous contexts of the reality, in addition to that the tests
false non match rates. Note, that the EER of a system can be can be repeated by others, and/or there can be a comparison at
read off at the point where its curve intersects with the diagonal the output level, e.g., recognition performance, without an im-
line. As can be observed in Fig. 9, the commercial matcher plementation. If one had the ground truth (de-noised images or
benefits from the pre-enhanced images in that the drop-off in minutiae positions and angles) for a database, additional testing
error rates is significant and occurs at all relevant operating strategies would have been possible. Unfortunately, there are no
points across all the four datasets. such datasets available because annotated databases are difficult
In our third experiment, we have studied the performance of to acquire.
the proposed minutiae extraction. To be precise, we have com- Beside the datasets used, we had a set of rolled fingerprints
pared the minutiae extracted by the proposed method with those out of NIST’s special database 14 at our disposal. After a
of NIST FIS2 (mindtct), quantitatively. We used the fingerprints rough adaptation to the ridge frequency of the fingerprints, the
enhanced by the introduced method from our first experiment proposed methods led to higher recognition rates than NFIS2
above, as they led to the best recognition results using the in- in initial experiments. The fingerprint pattern area is naturally
dependent NIST matcher. Feeding the minutiae, as extracted larger in case of rolled fingerprints and, thus, contains a lot


more minutia points, which increases the recognition potential by our strategy. If we additionally replace its minutiae extrac-
substantially. Also in terms of impression quality, the FVC tion with the proposed one, the error rates are decreased even
2004 is regarded as more challenging than all earlier collected further. Similarly, we have evaluated a commercial matcher
fingerprint databases. Since we wanted to show the benefits of in conjunction with the suggested enhancement method. The
our algorithms on more difficult fingerprint data, we did not results support that even state-of-the-art fingerprint recognition
give an account of performance with rolled fingerprints. systems benefit from the proposed techniques.
Our experiments showed that even a commercial grade
matcher can benefit from pre-enhancement. We conclude ACKNOWLEDGMENT
that this is only possible if the latter truly preserves signal The authors would like to thank the reviewers for their sug-
fidelity—in other words, if the enhancement does not distort gestions on experimental evaluation.
the fingerprint’s minutiae or texture. The reason for why the
other enhancement methods did not perform equally well is REFERENCES
not that they cannot deliver a meaningful enhancement as it is [1] A. K. Jain, S. Pankanti, S. Brabhakar, and A. Ross, “Recent advances
beyond doubt that their ability has been well demonstrated in in fingerprint verification,” in Audio and Video-Based Biometric Person
Authentication, J. Bigun and F. Smeraldi, Eds. New York: Springer,
published studies. However, it is to the best of our knowledge 2001, pp. 182–191.
the first time that they have been tested on a highly corrupted [2] D. Maltoni, D. Maio, A. K. Jain, and S. Prabhakar, Handbook
database which poses greater challenges on the underlying of Fingerprint Recognition. New York: Springer, 2003, includes
DVD-ROM.
assumptions and side effects of the image processing. [3] H. Fronthaler, K. Kollreider, and J. Bigun, “Automatic image quality
A benefit of the proposed techniques is that they are pos- assessment with application in biometrics,” in Proc. IEEE Workshop
sible to implement via fast signal processing techniques (1-D fil- on Biometrics, New York, Jun. 2006, pp. 30–35.
[4] E. Tabassi, C. Wilson, and C. Watson, “Fingerprint image quality,”
tering, pyramidal processing, etc.,) applied directly to the orig- Tech. Rep. NISTIR7151, 2004.
inal grayscale images, avoiding morphological operations all to- [5] Y. Chen, S. Dass, and A. Jain, “Fingerprint quality indices for pre-
gether. The latter also adds not the least to the accurateness of dicting authentication performance,” in Proc. Audio- and Video-based
Biometric Person Authentication, Jul. 2005, pp. 160–170.
the minutia extraction. Finally, our local features could also pro- [6] L. Hong, Y. Wand, and A. Jain, “Fingerprint image enhancement: Al-
vide added value (complementary information) in nonminutiae gorithm and performance evaluation,” IEEE Pattern Anal. Mach. In-
based fingerprint recognition systems. The parabolic symme- tell., vol. 20, no. 8, pp. 777–789, Aug. 1998.
[7] B. G. Sherlock, D. M. Monro, and K. Millard, “Fingerprint enhance-
tries reinforce the minutia definition and the linear symmetries ment by directional fourier filtering,” Vis. Image Signal Process., vol.
describe the texture outside of the minutia area, useful for ori- 141, no. 2, pp. 87–94, 1994.
entation matching in texture based fingerprint analysis. [8] S. Chikkerur and V. Govindaraju, “Fingerprint image enhancement
using STFT analysis,” in Proc. Int. Workshop on Pattern Recognition
for Crime Prevention, Security and Surveillance, 2005, pp. 20–29.
[9] C. I. Watson, G. T. Candela, and P. J. Grother, “Comparison of FFT
VIII. CONCLUSION fingerprint filtering methods for neural network classification,” NISTIR,
vol. 5493, 1994.
Symmetry features for local fingerprint image processing [10] A. Willis and L. Myers, “A cost-effective fingerprint recognition
have been presented and exploited in a novel image enhance- system for use with low-quality prints and damaged fingerprint,”
ment procedure as well as for reliable minutiae extraction. The Pattern Recognit., vol. 34, no. 2, pp. 255–270, Feb. 2001.
[11] S. Chikkerur, C. Wu, and V. Govindaraju, “A systematic approach for
enhancement is applied progressively, i.e., blockwise opera- feature extraction in fingerprint images,” in Proc. Int. Conf. Bioinfor-
tions are avoided, in the spatial domain. It does naturally not matics and its Applications, 2004, pp. 344–350.
suffer from blocking artifacts. Both absolute frequency (gran- [12] A. Almansa and T. Lindeberg, “Fingerprint enhancement by shape
adaptation of scale-space operators with automatic scale-selection,”
ularity or isotropic frequency) and orientation (nonisotropic IEEE Trans. Image Process., vol. 9, no. 12, pp. 2027–2042, Dec. 2000.
information) of the fingerprint pattern are utilized to obtain [13] H. F. D. Kunz, K. Eck, and T. Aach, “A nonlinear multi-resolution
the enhancement. The former is implemented by exploiting gradient-adaptive filter for medical images,” in Proc. SPIE Medical
Imaging, 2003, vol. 5032, pp. 732–742.
several levels of a bandpass pyramid and treating them inde- [14] D. Kaji, “Improvement of diagnostic image quality using a frequency
pendently. The typical ridge-valley flow is coherence enhanced processing based on decomposition into multiresolution space -hybrid
by using directional averaging and the structure tensor direction processing,” MI Solution Group, Tech. Rep., 2002.
[15] E. H. Adelson, C. H. Anderson, J. R. Bergen, P. J. Burt, and J. M.
(linear symmetry features) at each level. The processing of the Ogden, “Pyramid methods in image processing,” RCA Eng., vol. 29,
lowest level adds to the fidelity and details (conservation of no. 6, pp. 33–41, 1984.
minutiae) whereas the rough ridge-valley flow is cleaned and [16] E. P. Simoncelli and W. T. Freeman, “The steerable pyramid: A flexible
architecture for multi-scale derivative computation,” in Proc. Int. Conf.
gaps are closed at higher levels. Furthermore, the suggested Image Processing, Washington, DC, Oct. 1995, vol. 3, pp. 23–26.
minutiae extraction handles both ridge bifurcation and endings [17] J. Bigun and G. H. Granlund, “Optimal orientation detection of
and employs features from the enhancement, with addition linear symmetry,” in Proc. 1st Int. Conf. Computer VisionLondon,
Washington, DC, 1987, pp. 433–438.
of parabolic symmetry features. They allow for simultaneous [18] J. Bigun, Vision With Direction. New York: Springer, 2006.
detection of the minutiae’s position and direction. One-dimen- [19] H. Fronthaler, K. Kollreider, and J. Bigun, “Local feature extraction in
sional filtering is applicable for all tasks. The suggested local fingerprints by complex filtering,” in Proc. Int. Workshop on Biometric
Recognition Systems, Beijing, China, Oct. 22–23, 2005, vol. 3781, pp.
features can provide ridge-valley texture information also to 77–84.
nonminutiae based systems. Comprehensive test results on [20] A. Jain, L. Hong, and R. Bolle, “On-line fingerprint verification,” IEEE
challenging datasets (FVC2004) are favorable to the suggested Pattern Anal. Mach. Intell., vol. 19, no. 4, pp. 302–314, Apr. 1997.
[21] D. Maio and D. Maltoni, “Direct gray-scale minutiae detection in fin-
methods: A standardized minutiae matcher from NIST achieves gerprints,” IEEE Pattern Anal. Mach. Intell., vol. 19, no. 1, pp. 27–40,
the lowest equal error rate if the fingerprints are pre-enhanced Jan. 1997.


[22] M.-T. Leung, W. Engeler, and P. Frank, “Fingerprint image processing [38] R. Cappelli, D. Maio, and D. Maltoni, “Synthetic fingerprint-image
using neural networks,” in Comput. Commun. Syst., Sep. 24–27, 1990, generation,” presented at the Int. Conf. Pattern Recognition, 2000.
pp. 582–586. [39] N. K. Ratha and R. Bolle, Automatic Fingerprint Recognition Sys-
[23] K. Nilsson and J. Bigun, “Localization of corresponding points in fin- tems. New York: SpringerVerlag, 2003.
gerprints by complex filtering,” Pattern Recognit. Lett., vol. 24, pp.
2135–2144, 2003. Hartwig Fronthaler received the M.S. degree from
[24] D. Maio, D. Maltoni, R. Cappelli, J. Wayman, and A. Jain, “FVC the Department of Computer Systems Engineerin,
2004: Third fingerprint verification competition,” in Proc. Int. Conf. Halmstad University, Sweden.
Biometric Authentication, Hong Kong, Jul. 2004, pp. 1–7. His specialization has been in the field of image
[25] Home Page of NIST, [Online]. Available: http://www.itl.nist.gov/ analysis with a focus on biometrics. He joined the
div894/894.01/online.htm signal analysis group of Halmstad University in 2004.
[26] C. I. Watson, M. D. Garris, E. Tabassi, C. L. Wilson, R. M. McCabe, His major research interests are in the field of fin-
and S. Janet, “User’s guide to fingerprint image software 2–NFIS 2,” gerprint processing, including automatic quality as-
NIST, 2004. sessment and feature extraction. Furthermore, he has
[27] J. Bigun, T. Bigun, and K. Nilsson, “Recognition by symmetry deriva- been involved in research on face biometrics.
tives and the generalized structure tensor,” IEEE Pattern Anal. Mach.
Intell., vol. 26, no. 12, pp. 1590–1605, Dec. 2004.
[28] J. Bigun, “Recognition of local symmetries in gray value imagesby har-
monic functions,” in Proc. 9th Int. Conf. Pattern Recognition, Rome,
Italy, Nov. 14–17, 1988, pp. 345–347. Klaus Kollreider received the M.S. degree in com-
[29] H. Knutsson, M. Hedlund, and G. H. Granlund, “Apparatus for deter- puter systems engineering from Halmstad University,
mining the degree of consistency of a feature in a region of an image Sweden, in 2004, after which he joined its signal anal-
that is divided into discrete picture elements,” U.S. patent, 4.747.152, ysis group.
1988. His scientific interests include signal analysis
[30] J. Bigun, H. Fronthaler, and K. Kollreider, “Assuring liveness in and computer vision, in particular, face biometrics
biometric identity authentication by real-time face tracking,” in Proc. and anti-spoofing measures by object detection
IEEE Int. Conf. Computational Intelligence for Homeland Security and tracking. He is involved in a European project
and Personal Safety, Venice, Italy, Jul. 21–22, 2004, pp. 104–112. focused on biometrics, BioSecure, where he has
[31] M. Kass and A. Witkin, “Analyzing oriented patterns,” Comput. Vis, also contributed to a reference system for fingerprint
Graph., Image Process., vol. 37, pp. 362–385, 1987. matching.
[32] K. Nilsson and J. Bigun, “Using Linear Symmetry Features as a Pre-
Procesing Step for Fingerprint Images,” in Audio and Video Based
Person Authentication-VBPA 2001, J. Bigun and F. Smeraldi, Eds.
New York: Springer, 2001, pp. 247–252. Josef Bigun (M’88–SM’98) recieved the M.S. and
[33] K. Nilsson, “Symmetry filters applied to fingerprints,” Ph.D. disserta- Ph.D. degrees from Linköping University, Sweden,
tion, Chalmers Univ. Technol., Göteborg, Sweden, 2005. in 1983 and 1988, respectively.
[34] J. Bigun, G. H. Granlund, and J. Wiklund, “Multidimensional orienta- Between 1988 and 1998, he was with the EPFL,
tion estimation with applications to texture analysis and optical flow,” Switzerland. He was elected Professor to the Signal
IEEE Pattern Anal. Mach. Intell., vol. 13, no. 8, pp. 775–790, Aug. Analysis Chair at Halmstad University and the
1991. Chalmers University of Technology in 1998. His
[35] J. Fierrez-Aguilar, L.-M. Munoz-Serrano, F. Alonso-Fernandez, and J. scientific interests include a broad field in computer
Ortega-Garcia, “On the effects of image quality degradation on minu- vision, texture and motion analysis, biometrics,
tiae- and ridge-based automatic fingerprint recognition,” presented at and the understanding of biological recognition
the IEEE Int. Conf. Security Technology, Oct. 2005. mechanisms.
[36] B. Johansson, “Multiscale curvature detection in computer vision,” Dr. Bigun has cochaired several international conferences. He has been con-
M.S. thesis, Linköping Univ., Linköping, Sweden, 2001. tributing as a referee or as an editorial board member of journals including Pat-
[37] R. Cappelli, D. Maio, D. Maltoni, J. L. Wayman, and A. K. Jain, “Per- tern Recognition Letters and the IEEE TRANSACTIONS ON IMAGE PROCESSING.
formance evaluation of fingerprint verification systems,” IEEE Pattern He served on the executive committees of several associations, including IAPR.
Anal. Mach. Intell., vol. 28, no. 1, pp. 3–18, Jan. 2006. He has been elected Fellow of IAPR.

Fingerprint _prem

More Related Content

What's hot (18)

Similar to Fingerprint _prem (20)

Recently uploaded (20)

Fingerprint _prem