ISO/IEC TR 29794-4:2010
(Main)Information technology — Biometric sample quality — Part 4: Finger image data
Information technology — Biometric sample quality — Part 4: Finger image data
For aspects of quality specific to the finger image modality, ISO/IEC TR 29794-4:2010: specifies terms and definitions that are useful in the specification, use, and test of finger image quality metrics; defines the interpretation of finger image quality scores; identifies or defines finger image corpora for the purpose of serving as information for algorithm developers and users; develops statistical methodologies specific to finger image corpora for characterizing quality metrics to facilitate interpretation of scores and their relation to matching performance. Performance assessment of quality algorithms and standardization of quality algorithms are outside the scope of ISO/IEC TR 29794-4:2010.
Technologies de l'information — Qualité d'échantillon biométrique — Partie 4: Données d'image de doigt
General Information
Relations
Standards Content (Sample)
TECHNICAL ISO/IEC
REPORT TR
29794-4
First edition
2010-04-01
Information technology — Biometric
sample quality —
Part 4:
Finger image data
Technologies de l'information — Qualité d'échantillon biométrique —
Partie 4: Données d'image de doigt
Reference number
ISO/IEC TR 29794-4:2010(E)
©
ISO/IEC 2010
---------------------- Page: 1 ----------------------
ISO/IEC TR 29794-4:2010(E)
PDF disclaimer
This PDF file may contain embedded typefaces. In accordance with Adobe's licensing policy, this file may be printed or viewed but
shall not be edited unless the typefaces which are embedded are licensed to and installed on the computer performing the editing. In
downloading this file, parties accept therein the responsibility of not infringing Adobe's licensing policy. The ISO Central Secretariat
accepts no liability in this area.
Adobe is a trademark of Adobe Systems Incorporated.
Details of the software products used to create this PDF file can be found in the General Info relative to the file; the PDF-creation
parameters were optimized for printing. Every care has been taken to ensure that the file is suitable for use by ISO member bodies. In
the unlikely event that a problem relating to it is found, please inform the Central Secretariat at the address given below.
COPYRIGHT PROTECTED DOCUMENT
© ISO/IEC 2010
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail copyright@iso.org
Web www.iso.org
Published in Switzerland
ii © ISO/IEC 2010 – All rights reserved
---------------------- Page: 2 ----------------------
ISO/IEC TR 29794-4:2010(E)
Contents Page
Foreword .iv
Introduction.v
1 Scope.1
2 Normative references.1
3 Terms and definitions .1
4 Symbols and abbreviated terms .2
5 Finger Image Quality.2
5.1 Defect factors of finger image.2
5.2 Standardization approaches for exchange of finger image quality .3
6 Finger Image Quality Analysis .3
6.1 Introduction.3
6.2 Local Analysis.3
6.2.1 Constituent of Local Analysis.3
6.2.2 Approaches to Local Analysis of Finger Image .3
6.3 Global Analysis.9
6.3.1 Constituent of Global Analysis .9
6.3.2 Approaches to Global Analysis of Finger Image .10
6.4 Unified Quality Score.12
6.4.1 Methodology for Combining Quality Metrics.12
6.4.2 Weighted Average.13
6.4.3 Pattern Classifier.13
6.4.4 Area Consideration.14
Bibliography.15
© ISO/IEC 2010 – All rights reserved iii
---------------------- Page: 3 ----------------------
ISO/IEC TR 29794-4:2010(E)
Foreword
ISO (the International Organization for Standardization) and IEC (the International Electrotechnical
Commission) form the specialized system for worldwide standardization. National bodies that are members of
ISO or IEC participate in the development of International Standards through technical committees
established by the respective organization to deal with particular fields of technical activity. ISO and IEC
technical committees collaborate in fields of mutual interest. Other international organizations, governmental
and non-governmental, in liaison with ISO and IEC, also take part in the work. In the field of information
technology, ISO and IEC have established a joint technical committee, ISO/IEC JTC 1.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of the joint technical committee is to prepare International Standards. Draft International
Standards adopted by the joint technical committee are circulated to national bodies for voting. Publication as
an International Standard requires approval by at least 75 % of the national bodies casting a vote.
In exceptional circumstances, the joint technical committee may propose the publication of a Technical Report
of one of the following types:
⎯ type 1, when the required support cannot be obtained for the publication of an International Standard,
despite repeated efforts;
⎯ type 2, when the subject is still under technical development or where for any other reason there is the
future but not immediate possibility of an agreement on an International Standard;
⎯ type 3, when the joint technical committee has collected data of a different kind from that which is
normally published as an International Standard (“state of the art”, for example).
Technical Reports of types 1 and 2 are subject to review within three years of publication, to decide whether
they can be transformed into International Standards. Technical Reports of type 3 do not necessarily have to
be reviewed until the data they provide are considered to be no longer valid or useful.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO and IEC shall not be held responsible for identifying any or all such patent rights.
ISO/IEC TR 29794-4, which is a Technical Report of type 2, was prepared by Joint Technical Committee
ISO/IEC JTC 1, Information technology, Subcommittee SC 37, Biometrics.
ISO/IEC 29794 consists of the following parts, under the general title Information technology — Biometric
sample quality:
⎯ Part 1: Framework
⎯ Part 4: Finger image data [Technical Report]
⎯ Part 5: Face image data [Technical Report]
iv © ISO/IEC 2010 – All rights reserved
---------------------- Page: 4 ----------------------
ISO/IEC TR 29794-4:2010(E)
Introduction
The quality of finger image data is defined to be the predicted behavior of the image in a matching
environment. Thus, the quality information is useful in many applications. ISO/IEC 19784-1 and
ISO/IEC 19785-1 do allocate a quality field and specify the allowable range for the scores, with the
recommendation that the score be divided into four categories with a qualitative interpretation for each
category. Image quality fields are also provided in the fingerprint data interchange formats standardized in
ISO/IEC 19794-2, ISO/IEC 19794-3, ISO/IEC 19794-4, and ISO/IEC 19794-8. However, there is no standard
way to interpret the quality score that facilitates the interpretation and interchange of the finger image quality
scores.
The purpose of this part of ISO/IEC 29794 is to provide an informative technical report on methodologies for
objective, quantitative quality score expression and interpretation for finger images. It will complement
ISO/IEC 29794-1 in developing a reference finger image corpus. Such a reference corpus can be built upon
the availability of public finger images, which should then be used for quality score normalization.
© ISO/IEC 2010 – All rights reserved v
---------------------- Page: 5 ----------------------
TECHNICAL REPORT ISO/IEC TR 29794-4:2010(E)
Information technology — Biometric sample quality —
Part 4:
Finger image data
1 Scope
For aspects of quality specific to the finger image modality, this part of ISO/IEC 29794:
⎯ specifies terms and definitions that are useful in the specification, use, and test of finger image quality
metrics;
⎯ defines the interpretation of finger image quality scores;
⎯ identifies or defines finger image corpora for the purpose of serving as information for algorithm
developers and users;
⎯ develops statistical methodologies specific to finger image corpora for characterizing quality metrics to
facilitate interpretation of scores and their relation to matching performance.
Performance assessment of quality algorithms and standardization of quality algorithms are outside the scope
of this part of ISO/IEC 29794.
2 Normative references
The following referenced documents are indispensable for the application of this document. For dated
references, only the edition cited applies. For undated references, the latest edition of the referenced
document (including any amendments) applies.
ISO/IEC 29794-1, Information technology — Biometric sample quality — Part 1: Framework
3 Terms and definitions
For the purposes of this document, the terms and definitions given in ISO/IEC 29794-1 and the following apply.
3.1
foreground region
region of a finger image that contains valid finger image patterns
NOTE The most evident structural characteristic of a valid finger image is a pattern of interleaved ridges and valleys.
3.2
local region
block of m x n pixels of the foreground of a finger image, where m and n are smaller than the width and the
height of the finger image
3.3
finger image quality assessment algorithm
algorithm that reports a quality score for a given finger image sample
© ISO/IEC 2010 – All rights reserved 1
---------------------- Page: 6 ----------------------
ISO/IEC TR 29794-4:2010(E)
3.4
finger image corpus
collection of finger image samples
3.5
finger image quality category
common attribute or property of a group of finger images that causes them to perform or behave similarly for a
class of fingerprint matchers
4 Symbols and abbreviated terms
FQAA finger image quality assessment algorithm
DFT discrete Fourier Transform
QSN quality score normalization
QAID quality algorithm identification
ppi pixel per inch, which is analogous to dot per inch (dpi).
5 Finger Image Quality
5.1 Defect factors of finger image
A finger image obtained from a scanner is not always perfect. It may contain defects caused by the user
character (e.g user's skin condition), user behavior (e.g. improper finger placement), imaging (e.g scanner
limitation or imperfection), or environment (e.g. impurities on the scanner surface). Some of the defects and
their factors can be listed as follows:
1. Defect caused by user character
A. Extreme skin conditions such as very wet, very dry, etc.
B. Scars
C. Wrinkles
D. Blisters
E. Eczema
F. Impurities such as dirt, latent print, etc.
2. Defect caused by imaging
A. Sampling error
B. Low contrast or signal-to-noise ratio
C. Distortion
D. Erroneous or streak lines
E. Uneven background
F. Insufficient dynamic range
G. Non-linear or non-uniform grayscale output
H. Pixels not available due to hardware failure
I. Aliasing problems
3. Defect caused by user behavior
A. Elastic deformation
B. Improper finger placement such as too low, rotated, etc.
C. Insufficient area of finger image
4. Defect caused by environment
A. Humidity
B. Light
C. Impurities on the scanner surface
2 © ISO/IEC 2010 – All rights reserved
---------------------- Page: 7 ----------------------
ISO/IEC TR 29794-4:2010(E)
The performance of an automated fingerprint recognition system will be affected by the amount of defects or
the degree of imperfection present in the finger image. Therefore, it is necessary to compute the quality score
of the finger image produced. Section 6 suggests several possible approaches to compute the finger image
quality. The quality score shall be predictive of the performance of an automatic fingerprint recognition system.
Furthermore, the quality score should preferably be scanner-independent and source-independent.
5.2 Standardization approaches for exchange of finger image quality
As the finger image quality affects the performance of the fingerprint recognition system, the knowledge of
quality can and is currently being used to process finger images differently, by for example, invoking some
image enhancement methods prior to feature extraction, invoking different matchers based on quality or
simply changing the threshold of the system. In fact, the use of finger image quality to enhance the overall
performance of the system is increasingly growing. Therefore, there is a need to standardize the quantitative
quality score expression and interpretation so that a common interpretation of the quality scores is achieved.
This can be done, as suggested in ISO/IEC 29794-1, by either Quality Algorithm Identification (QAID), or
Quality Percentile Rank upon standardization of a Quality Score Normalization (QSN) corpus.
6 Finger Image Quality Analysis
6.1 Introduction
A complete finger image quality analysis should examine both the local and global structures of the finger
image. Fingerprint local structure constitutes the main texture-like pattern of ridges and valleys within a local
region while valid global structure puts the ridges and valleys into a smooth flow for the entire fingerprint. The
quality of a finger image is determined by both its local and global structures. This section describes the
current most significant features and characteristics of finger images at both local and global structures that
are related to performance of fingerprint recognition systems. Some of these algorithms are described in 6.2
and 6.3 and can also be found in [5-8,10,11].
The finger image is assumed to have resolution of 500 ppi. For other resolutions, the resolution dependent
parameters should be scaled accordingly. Possible initial finger image corpuses are the publicly available
Fingerprint Verification Competition (FVC) 2000, 2002, 2004, and 2006 [4] corpuses.
6.2 Local Analysis
6.2.1 Constituent of Local Analysis
A finger image is partitioned into blocks such that each block contains sufficient ridge-valley information,
preferably having at least 2 clear ridges, while not overly constraining the high curvature ridges. For images
with a resolution of 500 ppi, the ridge separation usually varies between 8 to 12 pixels [2]. A ridge separation
comprises a ridge and a valley. In order to cover two clear ridges, the block size has to be bigger than 24
pixels. Thus the suggested size for each block is 32 x 32 pixels, which is sufficient to cover 2 clear ridges.
Nevertheless, other sizes could also be used. Instead of Cartesian coordinate, curvilinear coordinate along the
ridge can also be used. This is followed by a segmentation process where each block is tagged as
background or foreground. There are several segmentation approaches, such as using the average
magnitude of the gradient in each block etc [2]. Local quality analysis is performed on the foreground blocks
with a local quality metric computed for each of them.
6.2.2 Approaches to Local Analysis of Finger Image
This section reviews some of the existing approaches for determining aspects of local quality of the finger
image.
© ISO/IEC 2010 – All rights reserved 3
---------------------- Page: 8 ----------------------
ISO/IEC TR 29794-4:2010(E)
6.2.2.1 Orientation Certainty Level
The finger image within a small block (as shown in Figure 1) generally consists of dark ridge lines separated
by white valley lines along the same orientation. The consistent ridge orientation and the appropriate ridge
and valley structure are distinguishable local characteristics of the fingerprint block.
Figure 1 — A typical texture-like ridge block
The grey level gradient (dx, dy) at a pixel describes the orientation and its strength at the pixel level. As an
example, [7] describes a method of measuring orientation certainty level. By performing Principal Component
Analysis on the image gradients in an image block, an orthogonal basis for an image block can be formed by
finding its eigenvalues and eigenvectors. Principal Components Analysis is a multivariate procedure which
rotates the data such that maximum variability is projected onto orthogonal axes. The resultant first principal
component contains the largest variance contributed by the maximum total gradient change in the direction
orthogonal to ridge orientation. The direction is given by the first eigenvector and the value of the variance
corresponds to the first eigenvalue, λ . On the other hand, the resultant second principal component has the
max
minimum change of gradient in the direction of ridge flow which corresponds to the second eigenvalue, λ .
min
The ratio between the two eigenvalues thus gives an indication of how strong the energy is concentrated
along the dominant direction with two vectors pointing to the normal and tangential direction of the average
ridge flow respectively. The covariance matrix C of the gradient vector for an N points image block is given by
1⎛⎞
⎡⎤dx ⎡a c⎤
Cd==⎡⎤xdy . (1)
⎜⎟
∑ ⎣⎦
⎢⎥ ⎢ ⎥
dy c b
N
⎣⎦ ⎣ ⎦
⎝⎠
N
For the covariance matrix in (1), eigenvalues λ are given by:
22
()ab++(a−b) +4c
λ = (2)
max
2
22
()ab+−(a−b) +4c
λ = (3)
min
2
For a finger image block, orientation certainty level (ocl), or the ratio between λ and λ is then:
min max
22
()ab+−(a−b) +4c
λ
min
ocl== (4)
λ 22
max
()ab++(a−b) +4c
The range of the ocl value is between 0 and 1 as a,b>0. It gives an indication of how strong the energy is
concentrated along the ridge-valley orientation. The lower the value the stronger it is. The value of ocl can
then be used to indicate the quality of the finger image block. The orientation certainty level fails to predict
match-ability when there exist some marks or residual in the samples that have strong orientation strength,
such as those exhibited by latent prints left by the previous user.
6.2.2.2 Ridge-valley Structure
Good quality fingerprints exhibit clear ridge-valley structure. Thus the measure of the ridge-valley structure
clarity is a useful indicator of the quality of a fingerprint.
4 © ISO/IEC 2010 – All rights reserved
---------------------- Page: 9 ----------------------
ISO/IEC TR 29794-4:2010(E)
Figure 2 — Extraction of a local region and transformation to vertical aligned ridge pattern
6.2.2.2.1 Ridge-valley Structure Analysis
An example of methods assessing clarity of ridge and valleys is explained in [7]. To perform ridge-valley
structure analysis, the finger image is quantized into blocks, preferably of size 32 × 32 pixels. Inside each
block, an orientation line, which is perpendicular to the ridge direction, is computed. At the centre of the block
along the ridge direction, a 2-D vector V (slanted square in Figure 2) of smaller size than the block size, such
1
as with size 32 × 16 pixels is extracted and transformed to a vertical aligned 2-D Vector V . By using equation
2
(5), a 1-D Vector V , that is the average profile of V , can be calculated.
3 2
m
Vi(,j)
∑
2
j=1
Vi() = , i = 1.32 (5)
3
m
where m is the block height (16 pixels) and i is the horizontal index.
Figure 3 — Region Segmentation of Vector V
2
Once V has been calculated, linear regression (or least square fitting) is then applied to V to find the
3 3
parameter, called Determine Threshold (DT ). DT is the line positioned at the centre of the Vector V , and is
1 1 3
used to segment the image block into the ridge or valley region. Regions with grey level intensity lower than
DT are classified as ridges; else they are classified as valleys. The process of segmenting the fingerprint
1
region into ridge and valley using DT is shown in Figure 3. The top portion of Figure 3 shows the ridge
1
pattern. The gray scale distribution of the ridge pattern projected as a one dimensional cumulative intensity
profile is shown at the lower portion. The Y-axis is the intensity level, while the x-axis the cross section of the
ridge pattern. Each local block will have its own DT .
1
From the one-dimensional signal in Figure 3, several useful parameters are computed, such as valley
thickness and ridge thickness. Since good finger images cannot have ridges that are too close or too far apart,
thus the nominal ridge and valley thickness can be used as a measure of the quality of the finger image
captured. Similarly, ridges that are unreasonably thick or thin indicate that the finger image may not be
captured properly, such as pressing too hard or too soft, or the image is a residual sample. Thus, the finger
image quality can be determined by comparing the ridge and valley thickness to each of their nominal range of
values. Any value out of the nominal range may imply a bad quality ridge pattern. The ridge and valley
thickness values are dependent on the resolution of the fingerprint scanner. To normalize these values, a
factor is computed by dividing the scanner resolution with 125 ppi which is the minimum resolution permitted
in ISO/IEC 19794-4. To normalize the range of the thickness values, a pre-set maximum thickness is used.
© ISO/IEC 2010 – All rights reserved 5
---------------------- Page: 10 ----------------------
ISO/IEC TR 29794-4:2010(E)
With a scanner resolution of 500 ppi, the maximum ridge or valley thickness (W ) for a good finger image is
max
estimated at 20 pixels or 5 pixels for a 125 ppi scanner in the normalized case. The pre-set value of 20 pixel
for a 500 ppi scanner resolution is obtained from the median of the typical ridge separation of 8 to 12 pixels [2],
and assuming that any ridge separation will not exceed twice of the median value. This will ensure that the
pre-set value is indeed the maximum to limit the value of the normalized ridge and valley thickness between 0
and 1. The ridge thickness (W ) and valley thickness (W ) are then normalized as follows:
r v
W
r
NW = ; where W = 5 (6)
max
r
((Sc /125) *W )
max
W
v
NW = ; where W = 5 (7)
max
v
((Sc /125) *W )
max
where NW and NW are the normalized ridge and valley thickness respectively and Sc the scanner resolution.
r v
With the ridge and valley separated as above, a clarity test can be performed in each segmented rectangular
2-D region. Figure 4 shows a sample grey level distribution of the segmented ridge and valley. The
overlapping area is the region of potential misclassification since in this region, whether a pixel belongs to
ridge or valley cannot be accurately determined using DT . Hence, the area of the overlapping region can be
1
an indicator of the clarity of ridge and valley, subject to the ridge and valley thicknesses being within the
acceptable range.
Figure 4 — Distribution of Ridge and Valley
The following equations describe the calculation of the clarity score, where υ is the number of pixels in the
B
valley with intensity lower than DT (also known as "bad pixel" for valley), υ is the total number of pixels in the
1 T
valley region, ℜ is the number of pixels in the ridge with intensity higher than DT (also known as "bad pixel"
B 1
for ridge), ℜ is the total number of pixels in the ridge region.
T
α =υυ/ (8)
BT
β =ℜ /ℜ (9)
BT
6 © ISO/IEC 2010 – All rights reserved
---------------------- Page: 11 ----------------------
ISO/IEC TR 29794-4:2010(E)
(αβ+<)/ 2 if ((NW NW
⎧
vvmin vmax rmin r rmax
LCS = (10)
⎨
1 otherwise
⎩
where ∧ is the logical AND operator; α and β are the portion of bad pixels while (NW , NW ) and (NW ,
rmin rmax vmin
NW ) are the minimum and maximum values for the normalized ridge (NW ) and valley (NW ) respectively.
vmax r v
Hence, the Local Clarity Score (LCS) is the constrained average value of α and β with a range between 0 and
1.
For ridges with good clarity, both distributions should have a very small overlapping area and thus LCS is
small. The following factors affect the size of the total overlapping area:
a. Noise on ridge and valley
b. Water patches on the image due to wet finger
c. Incorrect orientation angle due to the effect of directional noise
d. Scar across the ridge pattern
e. Highly curved ridges
f. Ridge endings, bifurcations, delta and core points.
g. Incipient ridges, sweat pores and dots
Factors (a) to (c) are physical noise found in the image. Factors (d) to (g) are actual physical characteristics of
the fingerprint. Therefore, a small window, such as with size 32 × 16, is chosen to minimize the chance of
encountering too many distinct features in the same location.
6.2.2.2.2 Directional Contrast
[8] describes a method for measuring clarity of ridge-valley structure by measuring directional contrast. The
value for directional contrast, D, is obtained by measuring the contrast between the gray values in the ridges
and the valleys along the orientation of the ridge flow [8]. The underlying idea is that the region of good quality
shows high directional contrast, which means that the ridges and the valleys in a given finger image are well
separated with regard to gray values. The overall process to calculate D is described in reference [8], and the
equation is simplified as follows. In Equation (11), Σ ( i, j ) is the sum of the pixels that follow the same
k
orientation, θ, and ( i, j ) represent the pixel indices in a block of size NxN of a finger image.
D=−θθ (11)
max ortho
NN
where θθ==max{ (ij, ),k= 1,.8} , and θθ⊥
max kk∑∑∑ ortho max
ij==11
k = orientation index
6.2.2.3 Frequency Domain Analysis
The signature of a high quality sample is a periodic signal, which can be approximated either by a square
wave or a sinusoidal wave. In the frequency domain, an ideal square wave should exhibit a dominant
frequency with sideband frequency components (sinc function). A sinusoidal wave consists of one dominant
frequency and minimum components at other non-dominant frequencies. Thus, we are able to make use of
such information in identifying good or bad quality blocks. The existences of one dominant frequency as well
as the frequency of such dominant components are two main elements that are useful in quality determination.
For each block a signature along the ridge-valley (x) direction, centered at the centre of each block as shown
below can be computed. The signature will pass through the centre of the image block in the direction of x as
shown in Figure 5.
© ISO/IEC 2010 – All rights reserved 7
---------------------- Page: 12 ----------------------
ISO/IEC TR 29794-4:2010(E)
Figure 5 — Signature along x direction
In the coordinate system of (x, y) as shown above, the signature is computed as :
r
1
Tx() = Ix( ,k) (12)
∑
21r +
kr=−
where I(x, y) is the grey level at point (x, y); x is the index along x axis and the range −≤25 x≤ 26 is usually
sufficient to cover two ridge separations [2] (Note that the exact value is not critical so long as it can cover at
least one periodic cycle completely); r is the width along the y axis that the signature is computed from and a
typical range used is −<10 r< 10 to obtain sufficient average grey level representation along the y axis. The
exact value is not critical but should not be too high to ensure that the direction of the ridges in the block is
consistent or too low which may not provide sufficient robustness against noise.
(a) Good Quality (b) Good Quality (c) Bad Quality (d) Bad Quality
Figure 6 — Image blocks with their respective DFTs of the signatures along the ridge direction.
Figure 6 above shows four finger image blocks with varying quality and their Discrete Fourier Transform (DFT)
of the signatures derived. The vertical axis of the plot is the DFT value of T(x) and the horizontal axis is the
index of the x-axis. Bad quality image (c) can be easily identified by the existence of dominant frequency at
very low f
...
Questions, Comments and Discussion
Ask us and Technical Secretary will try to provide an answer. You can facilitate discussion about the standard in here.