ISO/IEC 29794-4:2017

Information technology - Biometric sample quality - Part 4: Finger image data

ISO/IEC 29794-4:2017 establishes - terms and definitions for quantifying finger image quality, - methods used to quantify the quality of finger images, and - standardized encoding of finger image quality, for finger images at 196,85 px/cm spatial sampling rate scanned or captured using optical sensors with capture dimension (width, height) of at least 1,27 cm × 1,651 cm.

Technologies de l'information — Qualité d'échantillon biométrique — Partie 4: Données d'image de doigt

General Information

• Status: Published
• Publication Date: 20-Sep-2017
• ICS: 35.040 - Information coding
• ICS: 35.240.15 - Identification cards. Chip cards. Biometrics
• Technical Committee: ISO/IEC JTC 1/SC 37 - Biometrics
• Drafting Committee: ISO/IEC JTC 1/SC 37/WG 3 - Biometric data interchange formats
• Current Stage: 9599 - Withdrawal of International Standard
• Start Date: 30-Sep-2024
• Completion Date: 30-Oct-2025

Relations

• Revised: ISO/IEC 29794-4:2024 - Information technology - Biometric sample quality - Part 4: Finger image data
• Effective Date: 06-Jun-2022
• Revises: ISO/IEC TR 29794-4:2010 - Information technology - Biometric sample quality - Part 4: Finger image data
• Effective Date: 22-Sep-2012

Overview

ISO/IEC 29794-4:2017 - "Information technology - Biometric sample quality - Part 4: Finger image data" specifies standardized methods to quantify and encode finger image quality for fingerprint images captured at 196.85 px/cm (500 ppi) using optical sensors with minimum capture dimensions of 1.27 cm × 1.651 cm. The standard defines terms, a suite of quality metrics (local and global), preprocessing rules, and standardized binary and XML encodings for exchanging fingerprint quality information. A reference implementation for the normative metrics is available (NFIQ2).

Key topics and technical requirements

• Scope & sampling: Targets finger images at 196.85 px/cm (500 ppi) and minimum capture area.
• Preprocessing / segmentation: Requires removal of near-constant white margins (pixel-intensity threshold T = 250) and foreground/background segmentation before computing metrics.
• Local regions: Finger images are partitioned into 32 × 32 pixel local blocks to capture ridge-valley detail (at least two ridges per block).
• Normative quality metrics (selected):
  • Orientation certainty level
  • Local clarity score
  • Frequency domain analysis (FDA) score
  • Ridge-valley uniformity and orientation flow
  • Minutiae counts and minutiae-based quality measures
  • Region-of-interest image mean and orientation coherence measures
• Non-normative metrics: Radial power spectrum, Gabor quality score, and other auxiliary measures.
• Unified quality score: Methodology for combining individual metrics into a single quality score using training and classification techniques.
• Encoding & conformance: Defines binary and XML encodings for quality records, quality algorithm identifiers, and conformance levels (aligned with ISO/IEC 19794-1 Levels 1–3).

Practical applications and users

ISO/IEC 29794-4:2017 is used by:

• Biometric system developers and vendors - to implement consistent fingerprint quality assessment and to improve capture workflows.
• Device manufacturers - to validate optical sensor capture dimensions and sampling rates and to tune internal quality checks.
• Enrollment and verification systems - for real-time quality gating, re-capture prompts, and template selection.
• Testing labs and certification bodies - to benchmark and certify capture devices and algorithms against standardized quality metrics.
• Forensics and research - for objective, repeatable measures of fingerprint image utility and for algorithm development (reference: NFIQ2).

Related standards

• ISO/IEC 29794-1 (framework for biometric sample quality)
• ISO/IEC 19794 series (fingerprint data interchange formats)
• ISO/IEC 2382-37 (biometrics vocabulary)

Keywords: ISO/IEC 29794-4:2017, finger image quality, fingerprint image quality, biometric sample quality, NFIQ2, quality metrics, fingerprint segmentation, biometric interoperability.

Standards Content (Sample)

INTERNATIONAL ISO/IEC
STANDARD 29794-4
First edition
2017-09
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 2017
© ISO/IEC 2017, Published in Switzerland
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized otherwise in any form
or by any means, electronic or mechanical, including photocopying, or posting on the internet or an intranet, without prior
written permission. Permission can be requested from either ISO at the address below or ISO’s member body in the country of
the requester.
ISO copyright office
Ch. de Blandonnet 8 • CP 401
CH-1214 Vernier, Geneva, Switzerland
Tel. [phone]
Fax [phone]
[e-mail]
www.iso.org
ii © ISO/IEC 2017 – All rights reserved

Contents Page
Foreword .iv
Introduction .v
1 Scope . 1
2 Normative references . 1
3 Terms, definitions, symbols and abbreviated terms . 1
4 Conformance . 2
5 Finger image quality metrics . 2
5.1 Overview . 2
5.1.1 General. 2
5.1.2 Constituent of local quality metrics . 3
5.1.3 Constituent of global quality metrics . 3
5.1.4 Image preprocessing . 3
5.1.5 Image examples . 5
5.2 Normative contributive quality metrics . 5
5.2.1 General. 5
5.2.2 Orientation certainty level . 5
5.2.3 Local clarity score . 7
5.2.4 Frequency domain analysis (FDA) score .11
5.2.5 Ridge valley uniformity .13
5.2.6 Orientation flow .14
5.2.7 MU .15
5.2.8 MMB .15
5.2.9 Minutiae count in finger image .16
5.2.10 Minutiae count in center of mass region .16
5.2.11 Minutiae quality based on local image mean .16
5.2.12 Minutiae quality based on local orientation certainty level .17
5.2.13 Region of interest image mean .17
5.2.14 Region of interest orientation map coherence sum .19
5.2.15 Region of interest relative orientation map coherence sum .20
5.2.16 Quality feature vector composition .20
5.3 Non-normative quality metrics .23
5.3.1 General.23
5.3.2 Radial power spectrum .23
5.3.3 Gabor quality score .25
5.4 Unified quality score .27
5.4.1 Methodology for combining quality metrics .27
5.4.2 Training method .27
6 Finger image quality data record .28
6.1 Binary encoding .28
6.2 XML encoding .29
6.3 Quality algorithm identifiers .30
Annex A (normative) Conformance test assertions .32
Annex B (informative) Factors influencing fingerprint image character .44
Annex C (informative) Area consideration .46
Bibliography .47
© ISO/IEC 2017 – All rights reserved iii

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.
The procedures used to develop this document and those intended for its further maintenance are
described in the ISO/IEC Directives, Part 1. In particular the different approval criteria needed for the
different types of ISO documents should be noted. This document was drafted in accordance with the
editorial rules of the ISO/IEC Directives, Part 2 (see www.iso.org/directives).
Attention is drawn to the possibility that some of the elements of this document may be the subject of
patent rights. ISO shall not be held responsible for identifying any or all such patent rights. Details of
any patent rights identified during the development of the document will be in the Introduction and/or
on the ISO list of patent declarations received (see www.iso.org/patents).
Any trade name used in this document is information given for the convenience of users and does not
constitute an endorsement.
For an explanation on the voluntary nature of standards, the meaning of ISO specific terms and
expressions related to conformity assessment, as well as information about ISO's adherence to the
World Trade Organization (WTO) principles in the Technical Barriers to Trade (TBT) see the following
URL: www.iso.org/iso/foreword.html.
This document was prepared by Joint Technical Committee ISO/IEC JTC 1, Information technology,
Subcommittee SC 37, Biometrics.
This first edition cancels and replaces ISO/IEC/TR 29794-4:2010, which has been technically revised to
become an International Standard.
A list of all parts in the ISO 29794 series can be found on the ISO website.
iv © ISO/IEC 2017 – All rights reserved

Introduction
This document specifies finger image quality metrics. A reference implementation of the normative
metrics is available at https://github.com/usnistgov/NFIQ2.
The quality of finger image data is defined to be the degree to which the finger image data fulfils
specified requirements for the targeted application. Thus, the quality information is useful in many
applications. ISO/IEC 19784-1 allocates a quality field and specifies the allowable range for the scores,
with a 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. This
document defines a standard way to calculate the finger image quality score that facilitates the
interpretation and interchange of the finger image quality scores.
© ISO/IEC 2017 – All rights reserved v

INTERNATIONAL STANDARD ISO/IEC 29794-4:2017(E)
Information technology — Biometric sample quality —
Part 4:
Finger image data
1 Scope
This document establishes
— terms and definitions for quantifying finger image quality,
— methods used to quantify the quality of finger images, and
— standardized encoding of finger image quality,
for finger images at 196,85 px/cm spatial sampling rate scanned or captured using optical sensors with
capture dimension (width, height) of at least 1,27 cm × 1,651 cm.
2 Normative references
The following documents are referred to in the text in such a way that some or all of their content
constitutes requirements 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 2382-37, Information technology — Vocabulary — Part 37: Biometrics
ISO/IEC 19794-1:2011, Information technology — Biometric data interchange formats — Part 1:
Framework
ISO/IEC 29794-1, Information technology — Biometric sample quality — Part 1: Framework
3 Terms, definitions, symbols and abbreviated terms
3.1 Terms and definitions
For the purposes of this document, the terms and definitions given in ISO/IEC 2382-37, ISO/IEC 29794-
1 and the following apply.
ISO and IEC maintain terminological databases for use in standardization at the following addresses:
— IEC Electropedia: available at http://www.electropedia.org/
— ISO Online browsing platform: available at http://www.iso.org/obp
3.1.1
foreground region
set of all pixels of a finger image that form valid finger image patterns
Note 1 to entry: The most evident structural characteristic of a valid finger image is a pattern of interleaved
ridges and valleys.
© ISO/IEC 2017 – All rights reserved 1

3.1.2
local region
block of m × n pixels of the foreground of a finger image, where m and n are smaller than or equal to the
width and the height of the finger image
3.1.3
finger image quality assessment algorithm
algorithm that reports a quality score for a given finger image
3.1.4
metric
quantification of a covariate using a prescribed method
3.1.5
covariate
variable or parameter that either directly, or when interacting with other covariates, affects fingerprint
recognition accuracy
3.2 Symbols and abbreviated terms
DFT Discrete Fourier Transform
I matrix of grey-level intensity values corresponding to the pixels of an image
S ridge valley signature of a local region V
V matrix of grey-level intensity values corresponding to the pixels of a local region
4 Conformance
A finger image quality assessment algorithm conforms to this document if it conforms to the normative
requirements of Clause 5.
A finger image quality record shall conform to this document if its structure and data values conform
to the formatting requirements of Clause 6 (finger image quality data record) and its quality values are
computed using the methods specified in 5.2, 5.3 and 5.4.
Conformance to normative requirements of Clause 6 fulfils Level 1 and Level 2 conformance as specified
in ISO/IEC 19794-1:2011, Annex A. Conformance to normative requirements of 5.2 and 5.4 is Level 3
conformance as specified in ISO/IEC 19794-1:2011, Annex A.
5 Finger image quality metrics
5.1 Overview
5.1.1 General
Clause 5 establishes metrics for predicting the utility of a finger image (5.2 and 5.3). Image quality
metrics from a single image are useful to ensure the acquired image is suitable for recognition.
A complete finger image quality analysis shall 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.
Clause 5 describes the features and characteristics of finger images at both local and global structures
that are to be used for quantifying finger image quality.
2 © ISO/IEC 2017 – All rights reserved

For applying the algorithms as described in 5.2 and 5.3, the finger image shall have a spatial sampling
rate of 196,85 pixels per centimetre (500 pixels per inch).
5.1.2 Constituent of local quality metrics
A finger image is partitioned into local regions such that each local region contains sufficient ridge-valley
information, preferably having at least 2 clear ridges, while not overly constraining the high curvature
ridges. For images with a spatial sampling rate of 196,85 pixel per centimetre (500 pixel per inch), the
[1]
ridge separation usually varies between 8 pixels to 12 pixels . A ridge separation comprises a ridge
and a valley. In order to cover two clear ridges, the local region size has to be greater than 24 pixels in
both width and height. The size for each local region shall be 32 × 32 pixels, which is sufficient to cover
2 clear ridges. Instead of Cartesian coordinate, curvilinear coordinate along the ridge can also be used.
5.1.3 Constituent of global quality metrics
A global quality metric should be computed over the whole image and assess the utility of fingerprint
characteristics in the image.
5.1.4 Image preprocessing
5.1.4.1 Description
A segmentation process follows where each local region is labelled as background or foreground.
There are several segmentation approaches, such as using the average magnitude of the pixel-intensity
[1]
gradient in each local region .
This document does not prescribe segmentation methods, but notes that performing segmentation
influences the computed scores. Constant or near constant areas of the input image shall be removed
according to 5.1.4.2 prior to computing quality using the metrics specified in 5.2 and 5.3.
5.1.4.2 Removal of near constant white lines in image
Prior to computing features, fingerprint images are cropped to remove white pixels on the margins.
Starting from the outer margins, rows and columns with average pixel intensity above 250 are removed.
Pixel intensities take values [0, 255] for an 8-bit gray scale image. As a first approximation of the region
of interest, image columns and rows which are near constant white background are removed. Using the
algorithm specified below, a fixed threshold is set for gray scale pixel intensity of T = 250 to obtain the
μ
image without near constant areas.
The algorithm is specified as:
a) For each row R in I, starting from the top
i
1) Compute the row arithmetic mean μ
row
2) On the first occurrence where μ ≤T set idxi=
row μ t
3) On the last occurrence where μ ≤T set idxi=
row μ b
b) For each column C in I, starting from the left
i
1) Compute the column arithmetic mean μ
col
2) On the first occurrence where μ ≤T set idxi=
col μ l
3) On the last occurrence where μ ≤T set idxi=
col μ r
c) Extract the region of interest as Î = I.roi(idx , idx , idx , idx )
l t r b
© ISO/IEC 2017 – All rights reserved 3

5.1.4.3 Foreground segmentation based on local standard deviation
For quality features which require a foreground mask to indicate regions containing the fingerprint an
algorithm using local standard deviation is adopted.
The algorithm is specified as:
a) Normalize I to zero mean and unit standard deviation to produce Î
b) For each local region V in Î
1) Compute the standard deviation of V as σ
V
2) Mark the corresponding local region in I as foreground if σ > 0,1
mask V
5.1.4.4 Computing the dominant ridge flow orientation for a local region from pixel-intensity
gradients
The dominant ridge flow orientation is determined by computing the pixel-intensity gradient
information and then determining the orientation of the principal variation axis.
The numerical gradient of the local region is determined using finite central difference for all interior
pixels in x-direction and y-direction
IIxy+11,,−−xy
() ()
f = (1)
x
IIxy,,+11−−xy
() ()
f = (2)
y
With f and f , the dominant ridge flow orientation, angle (V), is determined analytically using the
x y
sine and cosine doubled angle determined from the arithmetic means of the pixel-intensity gradient
covariances.
a= f (3)
x
b= f (4)
y
c= ff (5)
xy
ac
C = (6)
 
cb
 
dc=+ ab− + (7)
()
c
sinθ = (8)
d
ab−
cosθ = (9)
d
1 sinθ
−1
angle V = tan (10)
()
2 cosθ
4 © ISO/IEC 2017 – All rights reserved

5.1.5 Image examples
For algorithms operating in a block-wise manner the input image is subdivided into local regions
according to the overlay grid shown in Figure 1 b). The local region V(8,5) is used as example in local
processing and is marked up using a bold line. Figure 1 c) shows an enlarged view of V(8,5) and Figure 1
d) shows V(8,5) rotated according to its dominant ridge orientation computed using Formula (10).
a) Input finger image b) division into local c) enlarged view of d) V(8,5) rotated
regions V(8,5) according to its
dominant ridge
orientation as
determined using
Formula (10)
Figure 1 — Input image used — Examples of the processing of quality
5.2 Normative contributive quality metrics
5.2.1 General
5.2 specifies normative contributive finger image quality assessment algorithms.
5.2.2 Orientation certainty level
5.2.2.1 Description
[3]
The orientation certainty level (OCL) of a local region is a measure of the consistency of the
orientations of the ridges and valleys contained within this local region. The feature computes local
quality and operates in a block-wise manner.
The finger image within a 32 × 32 pixels local region [as shown in Figure 1 c)] 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 local region.
The pixel-intensity gradient (dx, dy) at a pixel describes the direction of the maximum pixel-intensity
change and its strength. By performing Principal Component Analysis on the pixel-intensity gradients
in a local region, an orthogonal basis for the local region can be formed by finding its eigenvalues and
eigenvectors. 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 .
max
On the other hand, the resultant second principal component has the minimum change of gradient in
the direction of ridge flow which corresponds to the second eigenvalue, λ . The ratio between the
min
two eigenvalues thus gives an indication of how strong the energy is concentrated along the dominant
© ISO/IEC 2017 – All rights reserved 5

direction with two vectors pointing to the normal and tangential direction of the average ridge flow
respectively.
5.2.2.2 Computing the eigenvalues and local orientation certainty
From the covariance matrix C [Formula (6)] the eigenvalues λ and λ are computed as
min max
ab+− ab− +4c
()
λ = (11)
min
ab++ ab− +4c
()
λ = (12)
max
which yields a local orientation certainty level
λ

min
10−>, if λ
 max
local
λ
Q = (13)
 max
OCL

0, otherwise

which is a ratio in the interval [0,1] where 1 is highest certainty level and 0 is lowest.
NOTE The orientation certainty level fails to predict match-ability when some marks or residual exist in the
samples that have strong orientation strength, such as those exhibited by latent prints left by the previous user.
5.2.2.3 OCL algorithm
For each local region V in I:
a) compute the pixel-intensity gradient of V with centered differences method [Formulae (1), (2)];
b) compute the covariance matrix C [Formula (6)];
local
c) compute the eigenvalues of C to obtain Q [Formulae (11), (12), (13)].
OCL
Figure 2 visualizes the processing steps.
a) Current local region with the ratio between local
b) Local quality scores Q for example
OCL
eigenvalues marked as ellipse
fingerprint image
Figure 2 — Processing steps of orientation certainty level quality algorithm
6 © ISO/IEC 2017 – All rights reserved

5.2.3 Local clarity score
5.2.3.1 Description
Good quality fingerprints exhibit clear ridge-valley structure. Thus, the local clarity score (LCS)
[4]
, which is the measure of the ridge-valley structure clarity, is a useful indicator of the quality of a
fingerprint. The feature computes local quality and operates in a block-wise manner.
To perform ridge-valley structure analysis, the foreground of the finger image is quantised into local
[3]
regions of size 32 × 32 pixels . Inside each local region, an orientation line, which is perpendicular
to the ridge direction, is computed. At the centre of the local region along the ridge direction, a local
region of size 32 × 16 pixels shall be extracted and transformed to a vertically aligned local region.
On S, the local region average profile, calculated in 5.2.3.4, a linear regression (or least square fitting) is
applied to determine the Determine Threshold (DT) which is a line positioned at the centre of the local
region V, and is used to segment the local region into the ridge or valley region. Regions with grey level
intensity lower than DT are classified as ridges. Otherwise, they are classified as valleys.
Since good finger images cannot have ridges that are too close or too far apart, 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. To normalize the range of
the thickness values, a pre-set maximum thickness is used. The maximum ridge or valley thickness
(W ) for a good finger image is estimated at 20 pixels for a 196,85 pixel per centimetre (500 pixel
max
per inch). The pre-set value of 20 pixel for a 196,85 pixel per centimetre (500 pixel per inch) scanner
[1]
spatial sampling rate is obtained from the median of the typical ridge separation of 8 to 12 pixels ,
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 with respect
r v
to the maximum thickness.
With the ridge and valley separated as above, a clarity test can be performed in each segmented
rectangular 2-D region.
For local regions with good clarity, the pixel-intensity distribution of ridges and the pixel-intensity
local
distribution of the valleys have a very small overlapping area and thus Q is high. The following
LCS
factors affect the size of the total overlapping area:
a) noise on ridge and valley;
b) water patches on the image due to wet fingers;
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.
© ISO/IEC 2017 – All rights reserved 7

5.2.3.2 Computing the ridge valley signature of a local region
Given the local region V the ridge valley signature S is obtained by
V xy,
()
∑
y=1
S x = (14)
()
where V(x, y) is the grey level at point (x, y); x is the index along x-axis.
5.2.3.3 Determining the proportion of misclassified pixels
Formulae (15) and (16) specify the calculation of α and which are the proportion of pixels misclassified
respectively as valley or ridge. v is the number of pixels in valley region with intensity lower than DT
B
and v is the total number of pixels in valley region. r is the number of pixels in the ridge region with
T B
intensity higher than DT and r is the total number of pixels in the ridge region.
T
v
B
α = (15)
v
T
r
B
β = (16)
r
T
5.2.3.4 Determining the normalized ridge and valley width
The normalized valley width W and the normalized ridge width W are determined
v r
W
v
W = (17)
v
 S 
max
W
 
 
W
r
W = (18)
r
 S 
max
W
 
 
where
S is the scanner spatial sampling rate in dpi;
max
W is the estimated ridge or valley width for an image with 49,21 pixel per centimetre
(125 pixel per inch) spatial sampling rate;
W and W are the observed valley and ridge widths.
v r
max
According to Reference [1], W = 5 is reasonable for 49,21 pixel per centimetre (125 pixel per
inch) spatial sampling rate. By extension, the denominator in Formula (17) and the denominator in
Formula (18) shall be 20 for a spatial sampling rate of 196,85 pixels per centimetre (500 pixels per inch).
5.2.3.5 Computing the local clarity score
local
The local quality score Q is the constrained average value of α and β with a range between 0 and 1.
LCS
nmin nmaxmn in nmax
 αβ+
if W < 1− , ( v vv ) ( r rr )

local
Q = 2 (19)

LCS

0, ootherwise

where
8 © ISO/IEC 2017 – All rights reserved

nmin nmin
are the minimum values for the normalized ridge and valley width;
W and W
r v
nmax nmax
W and W are the maximum values for the normalized ridge and valley width.
v v
nmin
W = (20)
r
W
r
nmax
W = (21)
r
W
r
nmin
W = (22)
v
W
v
nmin
W = (23)
v
W
v
local
NOTE Particular regions inherent in a fingerprint will negatively affect �Q . For example, ridge endings
LCS
and bifurcations or areas with high curvature such as those commonly found in core and delta points.
5.2.3.6 LCS algorithm
For each local region V in I:
a) rotate V such that dominant ridge flow is perpendicular to x-axis;
b) crop rotated V such that no invalid regions are included;
c) with V obtain the ridge-valley signature S (5.2.3.2);
d) determine DT using linear regression on S;
e) for each element S(x), set threshold T(x) of x being ridge or valley based on DT;
1,i f(STxx)( < )
f) classify columns in V as ridge (1) or valley (0) with P()x ={ ;
0, otherwise
g) determine ridge-valley transition vector C from P;
h) compute the vector W containing ridge and valley widths from C;
i) determine normalized ridge width and valley width W and W (5.2.3.4);
r v
j) determine the proportion of misclassified pixels α and β (5.2.3.3);
local
k) compute the local quality score Q (5.2.3.5).
LCS
© ISO/IEC 2017 – All rights reserved 9

Figure 3 visualizes the processing steps.
a) Crop of current local b) Average profile of local region c) Average local region profile
region with linear regression line
d) Binarisation mask with e) Pixels determined to be ridge f) Pixel misclassified as valley
ridge and valley regions based on mask based on the threshold
based on regression line
g) Same as e) h) Same as e) i) Local clarity scores
Key
X block x-index
Y mean intensity
Figure 3 — Processing steps of local clarity score algorithm
10 © ISO/IEC 2017 – All rights reserved

5.2.4 Frequency domain analysis (FDA) score
5.2.4.1 Description
Frequency domain analysis (FDA) computes local quality and operates in a block-wise manner. A one-
dimensional signature of the ridge-valley structure is extracted and the Discrete Fourier Transform
(DFT) is computed on the signature to determine the frequency of the sinusoid following the ridge-
[5]
valley structure .
The ridge-valley 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.
For each local region, a signature perpendicular to the dominant ridge flow orientation is computed.
The FDA described in 5.2.4 computes the one-dimensional signatures by performing averaging along
the ridge flow direction. The averaging process filters out noise along the ridge and valley flow and
provides a modelling of a smooth changing signal in a direction perpendicular to ridge flow.
5.2.4.2 Computing the local FDA quality score
The local quality score is computed by using Formula (24):
1, if FF==AAor
maxm1 ax
A


local
AA++C A
Q = (24)
()
 FFF
FDA
maxmax−+11max
, otherwise

A//2

A
∑ F
F=1

where
0,3 is the attenuation parameter C;
A is the amplitude at frequency index x.
local
The value of Q is set to 1 when the maximum frequency F amplitude occurs at index
max
FDA
F = A or F = A .
max 1
max A
5.2.4.3 FDA algorithm
For each local region V in I:
a) pad V with 2 pixel border;
b) rotate V with nearest neighbour interpolation such that dominant ridge flow is perpendicular
to x-axis;
c) crop V such that no invalid regions are included;
d) with V obtain the ridge-valley signature S (5.2.3.2);
e) compute the DFT of S to obtain the magnitude representation A;
f) discard the first component of A;
g) determine F as the index with the largest magnitude in A;
max
local
h) compute Q of V using A and F (5.2.4.2).
max
FDA
© ISO/IEC 2017 – All rights reserved 11

Figure 4 visualizes the processing steps.
a) Central area of input local region b) Ridge-valley profile
2 000
1 500
c 1 000
1/16  1/8   3/16  1/4   5/16  3/8   7/16
d
c) DFT of ridge-valley profile local
d) Map of Q
FDA
a
Mean intensity.
b
Block x-index.
c
Magnitude.
d
Cycle/pixel.
Figure 4 — Processing steps of FDA quality algorithm
12 © ISO/IEC 2017 – All rights reserved

5.2.5 Ridge valley uniformity
5.2.5.1 Feature description
[3]
Ridge valley uniformity (RVU) is a measure of the consistency of the ridge and valley widths . The
expectation for finger image with clear ridge and valley separation is that the ratio between ridge and
valley widths remains fairly constant throughout the finger image.
The ratio of ridge thickness to valley thickness should be constant and close to 1 throughout the
whole image for a good quality finger image. The feature computes local quality and operates in a
block-wise manner.
5.2.5.2 RVU algorithm
For each local region V in I:
a) determine dominant ridgeflow orientation angle (V) of V;
b) rotate V such that angle (V) is perpendicular to x-axis;
c) crop V such that no invalid regions are included;
d) with V obtain the ridge-valley signature S (5.2.3.2);
e) determine DT using linear regression on S;
f) for each S(x) compute threshold T(x) = x × DT(1) + DT(0);
g) binarize S using T;
11,i f(STxx−<)( )
h) classify ridge and valley in S as P()x ={ ;
0, otherwise
11,i f(PPxx−≠)( )
i) compute ridge-valley transition vector as C()x ={ ;
0, otherwise
j) Drop first and last transition from S using C to remove incomplete ridges or valleys and obtain S’;
local
k) Compute Q as the ratio between widths of ridge and valleys in S’.
FDA
Figure 5 visualizes the processing steps.
© ISO/IEC 2017 – All rights reserved 13

a) Crop of current local region b) Average profile of local region
c) Average profile with regression line d) Local quality score as the standard deviation
of local ridge to valley ratios
Key
X block x-index
Y mean intensity
Figure 5 — Processing steps of ridge valley uniformity quality algorithm
NOTE The ridge valley uniformity quality feature is spatial sampling rate dependent. The given defaults
assume 196,85 pixel per centimetre (500 pixel per inch).
5.2.6 Orientation flow
5.2.6.1 Description
[4]
Orientation flow (OFL) is a measure of ridge flow continuity which is based on the absolute orientation
difference between a local region and its 8-neighborhood of local regions.
Orientation flow is a good indicator to describe the quality of a good fingerprint pattern because, in
general, the flow of the ridge direction changes gradually, except in an area with a delta or a core. The
feature computes local quality and operates in a block-wise manner.
5.2.6.2 Local region-wise absolute orientation difference
The ridge flow is determined as a measure of the absolute difference between a local region and its
neighboring local regions. The absolute difference D(i, j) for local region V(i, j) is computed using the
dominant ridge flow orientations of this local region and of its neighbors
1 1
angleaVVij,,−−ngle im jn−
()() ()()
∑∑
mn=−1 =−1
D ij, = (25)
()
14 © ISO/IEC 2017 – All rights reserved

5.2.6.3 Local orientation flow quality score
local
The local orientation quality score Q for the local region orientation difference D(i, j) is
OFL
D ij, −θ
()
min
,,ifD ij >θ
 ()
min
local
90°−θ
Q = (26)

min
OFL

0, otherwise

where θ = 4 is the threshold for minimum angle difference to consider.
min
5.2.6.4 OFL algorithm
a) Determine the dominant ridge flow orientation angle (V) of local region V in I.
b) For each local region V in I:
1) compute the absolute orientation difference D(i, j) using angle (V) (5.2.6.2);
local
2) compute the local orientation quality score Q (5.2.6.3).
OFL
Figure 6 visualizes the processing steps.
a) Line marking the b) Local orientations c) Orientation d) Local quality scores
normal to the ridgeline differences
orientation
Figure 6 — Processing steps of orientation flow quality algorithm
5.2.7 MU
5.2.7.1 Description
The MU quality feature is the arithmetic mean of the pixel intensities of all pixels in the input image.
The feature computes global quality.
5.2.7.2 MU algorithm
Compute Q as the arithmetic mean of pixel intensities in I.
MU
5.2.8 MMB
5.2.8.1 Description
The MMB quality feature is the arithmetic mean of per local region computed arithmetic mean in the
gray scale input image. The feature computes local quality and operates in a block-wise manner.
© ISO/IEC 2017 – All rights reserved 15

5.2.8.2 MMB algorithm
a) For each local region V in I
local
1) compute the arithmetic mean of the pixel intensities in V as Q .
MMB
local
b) Compute Q as the arithmetic mean of set of Q .
MMB
MMB
5.2.9 Minutiae count in finger image
5.2.9.1 Description
The FingerJet FX (FJFX) minutiae extractor provides a count of detected minutiae in the finger image.
The minutiae count has a bearing on the mated comparison score. The feature computes global quality.
CNT
5.2.9.2 MIN algorithm
cnt
Q is the number of detected minutiae in the finger image as determined by FJFX.
MIN
5.2.10 Minutiae count in center of mass region
5.2.10.1 Description
The FingerJet FX (FJFX) minutiae extractor provides locations of detected minutiae in a finger image.
The feature is the minutiae count in a 200 × 200 pixels local region centered on the center of mass of the
detected minutia. The feature computes local quality at the minutiae locations.
COM
5.2.10.2 MIN algorithm
com
Q is the number of minutiae occurring within a 200 × 200 pixels local region centered at the center
MIN
of mass of the locations of all detected minutiae in the finger image as determined by FJFX.
5.2.11 Minutiae quality based on local image mean
5.2.11.1 Description
The FingerJet FX (FJFX) minutiae extractor provides locations of detected minutiae in a finger image.
For each minutia location a local quality based on image statistics is computed. The reported quality
value is aggregated as the count of local qualities which occurs in the specified range. The feature
computes local quality at the minutiae locations.
MU
5.2.11.2 MIN algorithm
mu
Q is computed by first determining the local quality of each minutiae detected by FJFX as
MIN
μμIV−
() ()
local
mu
Q = (27)
MIN
σ I
()
where μ(I) and μ(V) is arithmetic mean of respectively the finger image and a 32 × 32 pixels local region
centered on the minutia and σ(I) is the standard deviation of the finger image.
16 © ISO/IEC 2017 – All rights reserved

local
mu
mu
The minutiae quality feature Q is finally computed as the percentage of Q which have values
MIN
MIN
between 0 and 0,5 as
local
mu cnt
mu
QQ= xy,,|f00≤< 50, or <≤i Q (28)
()
MIN {} MIN
MIN
5.2.12 Minutiae quality based on local orientation certainty level
5.2.12.1 Description
The FingerJet FX (FJFX) minutiae extractor provides locations of detected minutiae in a finger image.
For each minutia location a local orientation certainty level is computed. The reported quality value is
aggregated as the count of local qualities which exceed the specified value. The feature computes local
quality at the minutiae locations.
OCL
5.2.12.2 MIN algorithm
ocl
Q is computed by first determining the local quality of each minutiae detected by FJFX as
MIN
local
local
ocl
QQ= V (29)
()
MIN OCL
local
where Q V is the local orientation certainty level (5.2.2) for the 32 × 32 pixels local region V
()
OCL
centered on the minutia.
local
ocl
ocl
The minutiae quality feature Q is finally computed as the percentage of Q which have values
MIN MIN
greater than 0,8 as
local
ocl cnt
ocl
QQ= xy,,|f>08 or 0<≤i Q (30)
()
MIN {} MIN
MIN
5.2.13 Region of interest image mean
5.2.13.1 Description
The region of interest for the finger image is the foreground region of the image containing the
fingerprint. The mean image intensity in this area is computed over the set of 32 × 32 pixels local
regions which have a least one pixel contained in the region of interest. The feature computes global
quality.
NOTE The quality score is highly correlated with Q (5.2.7) and Q (5.2.8).
MU MMB
5.2.13.2 AREA algorithm
a) Determine the region of interest R (5.2.13.3).
b) For each 32 × 32 local region V in I
1) if V has at least 1 pixel contained in foreground of R, mark the local region as foreground.
μμ
c) Compute Q as the arithmetic mean of the set of V which are marked as foreground.
AREA
5.2.13.3 Determine the Region of Interest
a) Erode the finger image I with 5 × 5 structuring element to obtain I’.
© ISO/IEC 2017 – All rights reserved 17

b) Apply normalized Gaussian blur filter (each weight is divided by the sum of all weights) with kernel
size 41 × 41 and standard deviation of 6,5 to I’ to obtain G.
[6]
c) Binarize G using Otsu’s method to obtain B.
d) Apply normalized Gaussian blur filter (each weight is divided by the sum of all weights) with kernel
size 91 × 91 and standard deviation of 14,0 to B to obtain G’.
e) Binarize G’ using Otsu’s method to obtain B’.
[7]
f) Determine the contours of B’ using Suzuki’s method to obtain C.
g) Regions in C which are surrounded by 0 valued pixels shall be set to 0 valued pixels.
h) 0
...