Li S.Z., Jain A.K. (eds.) Encyclopedia of Biometrics

Подождите немного. Документ загружается.

points in the lines connecting the detected edges. Oth-

erwise, we let the initial points lie in the detected

circle. In both cases, the snake points are initialized

densely in the solid lines but sparsely in the dashed

lines in the templates in Fig. 3; we assume that the

outer boundaries around the solid lines, i.e., the

left and right sides of the outer boundaries are more

robust on the occlusion by eyelids. As shown in

the ﬁnal conﬁguration of snakes in Fig. 3d, the

snake points lie in the boundary almost uniformly in

the end.

S ¼



20:5

0:51



ð4Þ

Iris Segmentation Using Active Contours. Figure 3 Segment an outer boundary at sclera. (a) the results of circle

detection; (b) two templates of outer boundaries; (c) initial points generated by either of the two templates; (d) ﬁnal

results of snakes.

Iris Segmentation Using Active Contours. Figure 2 Snakes at Kth iteration for detecting an inner boundary. (a) initial

points; (b) K =1;(c) K = 2; (d) K =5;(e) K = 12; (f) K = 30.

850

Iris Segmentation Using Active Contours

Segmentation Results

Figure 4 shows the results of iris segmentation pro-

posed in this chapter. For experiments, the NIST ICE-1

database, which was constructed for NIS T Iris Chal-

lenge Evaluation [8], is used. The segmentation results

demonstrate that the proposed method using snakes

produces reliable results in various cases. In particular,

the segmentation results show that the proposed

method is powerful to detect outer boundaries even

when upper eyelids and eyelashes make severe occlu-

sions inside iris regions. By active contours, the eyelids

are excluded and the noncircular bound aries are

detected successfully as shown in Fig. 4.

After the snake points converge, we can estimate

the iris boundaries roughly. Next, we obtain more

reliable boundaries by connecting the ﬁnal snake

points in an outer or inner boundary into a closed

line, and smoothing the connected line as shown in

Fig. 5.

Iris Segmentation Using Active Contours. Figure 4 Examples of iris boundary detection using snakes.

Iris Segmentation Using Active Contours. Figure 5 Segment a true iris region after applying snakes. (a) ﬁnal results of

snakes; (b) Smooth the boundary connecting the snake points; (c) ﬁnal iris region.

Iris Segmentation Using Active Contours

851

Summary

For successful iris recognition, it is indispensable to

segment the true iris region from an image. So, as a

reliable iris recognition technique is demanded, the

demand for robust iris segmentation also increases.

However, iris segmentation is challe nging in that

upper and lower eyelids and eyelashes often make

occlusion which is difﬁcult to be detected by previous

edge detectors or circle detectors. To improve the

results of iris segmentation, active contours called

snakes have been applied to iris boundary detection.

In particular, snakes provide a powerful segmentation

ability to detect and remove the occlusion by an upper

eyelid which has been one of the most signiﬁcant

obstacles for reliable iris segmentation.

Related Entries

▶ Iris Recognition

References

1. Li, S.Z., Jain, A.K. (eds.): Handbook of Face Recognition.

Springer, New York (2005)

2. Blake, A., Isard, M.: Active Contours. Springer, Heidelberg,

Germany (1988)

3. Kass, M., Witkin, A., Terzopoulos, D.: Snakes: Active Contour

Models 1, 321–331 (1988)

4. Bowyer, K., Hollingsworth, K., Flynn, P.: Image understanding

for iris biometrics: A survey. University of Nortre Dame CSE

Tech. rep., (2007)

5. Daugman, J.: High conﬁdence visual recognition of persons by a

test of statistical independence. IEEE Trans. Pattern Anal. Mach.

Intell. 15(11) (1993)

6. Daugman, J.: How iris recognition works. IEEE Trans. Circuits

Syst. Video Technol. 14(1), 21–30 (2004)

7. Thornton, J., Savvides, M., Vijaya Kumar, B.: A bayesian ap-

proach to deformed pattern matching of iris images. IEEE Trans.

Pattern Anal. Mach. Intell. 29(1), 596–606 (2007)

8. Park, S.: National Institute of Standards and Technology. Iris

Challenge Evaluation. http://iris.nist.gov/ice/

Iris Segmentation Using Snakes

▶ Iris Segmentation Using Active Co ntours

Iris Standards Evolution

▶ Iris Standards Progression

Iris Standards Progression

DOMINIQUE HARRINGTON

,RYAN TRIPLETT

Tygart Technology Inc., Fairmont, MV, USA

Biometric Services International, LLC, A National

Biometric Security Project Company, Morgantown,

WV, USA

Synonym

Iris standards evolution

Definition

Iris standards are used to provide a set of guidelines for

iris system implementation, design, and interopera-

bility. Both published and emerging standards for

iris-based systems include recommendations for inter-

operability, data formats, conformance, compression,

and quality.

There are two Technical Committees that create

standards for the United States (U.S.) and internatio-

nal communities which solely concern biometrics. The

American National Standards Institute (ANSI) Inter-

National Committee for Information Technology

Standards (INCITS) M1 Technical Committee creates

U.S. national standards. This same group is the U.S.

Technical Advisory Group for the International Orga-

nization for Standardization (ISO) and the Inter-

national Electrotechnical Commission (IEC) Joint

Technical Committee 1, Subcommittee 37 (SC37),

which creates biometric international standards

[1, 2]. Both ANS I/INCITS and ISO/IEC standards

bodies ultimately publish the iris standards that are

discussed within this chapter.

Introduction

Biometric standards development began in November

2001 with the creation of M1 in the United States.

852

Iris Segmentation Using Snakes

Shortly thereafter, the international community began

developing biometric standards with the creation of

SC37 in June 2002. Each technical committee is broken

into separate subcommittees. The overall structure of

both organizations is simil ar with only the subcom-

mittee names being different, as shown below:



M1.1/WG1 – Task Group on Biometric Vocabulary



M1.2/WG2 – Task Group on Biometric Technical

Interfaces



M1.3/WG3 – Task Group on Biometric Data Inter-

change Formats



M1.4/WG4 – Task Group on Biometric Proﬁles



M1.5/WG5 – Task Group on Biometric Perfor-

mance Testing and Reporting



M1.6/WG6 – Task Group on Cross Jurisdictional

and Societal Issues

These subcommittees have created several standards

governing the adoption of biometric technology

throughout the industry. However, there are a few

standards that speciﬁcally involve iris technology.

These standards have either been published, or are

emerging documents about to be published. The cur-

rent versions are as follows:



ANSI INCITS 379-2004 American National Stan-

dard for Informati on technology – Iris Interchange

Format



ANSI INCITS 1749-D: Part 6 -Conformance

Testing Methodology for INC ITS 379, Iris Image

Interchange Format



ISO/IEC 19794-6 Information Technolog y –

Biometric Data Interchange Format – Part 6: Iris

Image Data



ISO/IEC 29109-6 (Base Docum ent) Information

Technology – Confor mance Testing Methodology

for Biometric Data Interchange Records as deﬁned

in ISO/I EC 19794 Biometric Data Interchange

Format Standard – Part 6: Iris Image Data

The following sections will provide details concerning

the speciﬁc standards that have been drafted and the

evolution of standards development planned for the

years to come.

Data Format Standards

Currently, data format standards are the only two

standards published by both national and international

bodies. They are the ANSI INCITS 379 Iris Image Inter-

change Format and the ISO/IEC 19794-6 Information

Technology – Biometric Data Interchange Format – Part 6:

Iris Image Data. Both of these standards are used to

provide manufacturers an accepted data format to

increase interoperability between multiple iris systems.

These standards identify two image interchange

formats for biometric authentication systems that uti-

lize the biometric iris modality. The ﬁrst is

▶ rectilinear,

consisting of a raw image, or a compressed image using

the JPEG compression standard. The second is the

▶ polarized format, which requires pre-processing

and image segmentation.

In addition to the interchange formats there are

some speciﬁcations identiﬁed that are consistent

throughout both standards. The following section out-

lines these speciﬁcations in detail.

Data Format Specifications

Each record obtained must have an iris record header

which contains information about the image capture

device and conditions by which the image w as cap-

tured. The header will identify the number of

features recorded, each iris as captured from the left

or right, and the number of images captured from each

eye. Each iris record should also contain an image

sequence number with information about the quality

and rotation of the image. An iris image data record

will contain either rectilinear or polar format images

and should not be mixed.

All images should have a 256 gray level range,

allocating at least one byte per intensity value and

provide at least 7 bits of useful intensit y information.

▶ specularity reﬂections from the illumination

source occur, their intensity should be set to the satu-

ration level or to a gray value of zero.

The eye should be illuminated by using near-infrared

wavelengths of approximately 700 and 900 nm. The

angle between a line exten ding from the center of

the illumination source to the center of the pupil,

and the optical axis of the iris camera should be at

least 5



to prevent ‘‘red-eye’’ effect.

The iris image should have a minimum of 70 gray

levels of separation between the iris and the sclera, and

a minimu m of 50 gray levels of separation between

the iris and the pupil for any eye color. A minimum

of 70% of the iris should be visible and not obscured by

Iris Standards Progression

853

specular reﬂections or any other obstructions. Table 1

outlines the quality recommendations by both the

standards bodies while Table 2 discusses the deﬁnition

of minimum requirement, medium quality, and high

quality iris data records.

Conformance Standards

In accordance with the data format standards

published, both M1 and SC37 are rapidly developing

the conformance testing standards that will show com-

pliance with the existing published standards. The

standards are in their ﬁnal development stages and

are planned to be completed in 2008 [1, 2].

The conformance tests, in accordance with stan-

dards, produce veriﬁable results that a given manu-

facturer has designed a system capable of being

interoperable with other iris systems available on the

market. Such standards have also identiﬁed require-

ments for testing to maintain consistency.

Current Standard Bodies

Developments

Like any emerging technology, new developments

in the ﬁeld of iris recognition are constantly being

discovered, which have had di rect effects on the stan-

dards community. In May 2007, the University of

Cambridge released a technical report titled Effect of

severe image compression on iris recognition perfor-

mance [5]. This report documented three compression

methods that retained rectilinear image formats com-

pressing to as little as 2,000 bytes while still allowing

quality recognition performance on publicly available

iris image databases. The compression methods are as

follows:

The ﬁrst method reduced the size of the standar-

dized iris image format by cropping the image and

then compressed the cropped image into the JPEG

format. ‘‘The standardized format is 640 by 480 pixels

with 8 bits grayscale data per pixel, which uses approx-

imately 307,200 bytes. Results show that a reduction

Iris Standards Progression. Table 1 Quality recommendations outlined in standards

Image

Quality

level

Image

Quality

value

Expected iris

diameter, pixels

Minimum pixel

resolution, pixels

per mm

Optical resolution at 60%

modulation, lp/mm Comment

poor 0–25  Unacceptable

quality

low 26–50 100–149 8,3 2,0 Marginal

quality

medium 51–75 150–199 12,5 3,0 Acceptable

quality

high 76–100 200 or more 16,7 4,0 Good quality

Iris Standards Progression. Table 2 Minimum, Medium, and High Quality Defined

Quality Definition

Minimum

Requirement

Image quality values between 0 and 25 and are used to indicate the image does not meet the

minimum quality standards

Images identified as being low quality use cameras with a minimum spatial resolution of 2.0 lp/mm

at 60% or higher contrast, and pixel resolution at least 8.3 pixels per mm at the object plane [3, 4]

Medium Quality Image quality values in the range of 51–75 are considered to be of medium quality

Images identified as being medium quality use cameras with a spatial resolution of at least

3.0 lp/mm at 60% or higher contrast, and pixel resolution of at least 12.5 pixels per mm at the

object plane [3, 4]

High Quality Image quality values in the range of 76–100 are considered to be of high quality and should use

cameras which have a spatial resolution of at least 4.0 lp/mm at 60% or higher contrast

The pixel resolution should be at least 16.7 pixels per mm at the object plane [3, 4]

854

Iris Standards Progression

factor of 72:1 increased the Equal Error Rate (EER) by

a factor of .0013.’’

The second method reduced the size of the image

to an allocated Region of Interest (ROI). ‘‘This is done

by substituting all non-iris parts of the image with a

uniform grayscale; darker gray is used to signify

eyelashes and lighter gray is used to signify the sclera.

Highlighting the ROI at every Quality Factor has

proven to be beneﬁcial.’’

The third method reduced the data size using

JPEG2000 compression. ‘‘This method is similar to

the JPEG compression, but compresses the image

further by a facto r of 20–30%. JPEG2000 also allows

use of a mask to specify an arbitrary shape (ROI) to

control the allocation of the encoding budget. Results

showed that a reduction factor of 180:1 using

JPEG2000 compression and locating the ROI simulta-

neously increased the EER by .0016.’’

Publishing the technical report, moved SC37 WG3,

at the Berlin meeting in June 2007, to propose the

removal of the polar image format from the ISO/IEC

standard because of defects caused by polar image

distortion. M1 representatives were concerned about

the impact the change would cause in the U.S. national

biometric community. They recommended waiting

until additional research and analysis was complete

before making a ﬁnal decision regarding removal of

the polar format. The Department of Homeland

Security (DHS) backed this decision by submitting a

Position Paper in support of retaining the I ris Polar

Image Data Format to M1 [6]. The paper outlined the

effects on DHS’s Registered Traveler (RT) program and

general concerns with the technical report. As of

December 3 2007 M1 representatives will submit

their ballots, determining the acceptance or withdraw-

al of Project 1576-D – revision of INCITS 379–2004.

Future Standards

In addition to ANSI and ISO, there are other organiza-

tions that are concerned with biometric implemen-

tations both nationally and internationally. These

organizations have created overall speciﬁcations and

best practice documents that discuss biometrics as a

whole. In some cases, these groups work with M1 and

SC37 to establish future standards that meet the

requirements of cross functional organizations outside

the biometric industry mainstream. Table 3 identiﬁes

few of these organizations.

Summary

During the last six years, biometric standards have

progressed rapidly compared with similar technology

related industries from 6 in 2004 to the current 53 pub-

lished standards. The current emerging and published

Iris Standards Progression. Table 3 Other organizations affection standards bodies

Organization name Brief overview

X9F4 (Financial) The Accredited Standards Committee X9 (ASC X9) has the mission to develop,

establish, maintain, and promote standards for the Financial Services Industry in

order to facilitate delivery of financial services and products

International Labour Organization –

ILO

The International Labour Organization (ILO) is the tripartite UN agency that

brings together governments, employers, and workers of its member states in

common action to promote decent work throughout the world

Institute of Electrical and Electronics

Engineers – IEEE

The IEEE promotes the engineering process of creating, developing, integrating,

sharing, and applying knowledge about electro and information technologies

and sciences for the benefit of humanity and the profession

National Institute of Standards and

Technology – NIST

Founded in 1901, NIST is a non-regulatory federal agency within the U.S.

Commerce Department’s Technology Administration. NIST’s mission is to

develop and promote measurement, standards, and technology to enhance

productivity, facilitate trade, and improve the quality of life

National Physical Laboratory – NPL The National Physical Laboratory (NPL) is the UK’s National Measurement

Institute and is a world-leading centre of excellence in developing and applying

the most accurate measurement standards, science and technology

Iris Standards Progression

855

standards are proving to be only the beginning in

the development of standards. Memberships in both

groups are continuing to grow and future standards

development is imminent.

Related Entries

▶ Biometrics System Design

▶ Interoperable Performance

▶ Iris Device

▶ Iris Encoding and Recognition

▶ Iris Image Quality

▶ Performance Evaluation, Overview

References

1. http://isotc.iso.org

2. http://m1.incits.org/

3. ANSI INCITS 379-2004 Iris Image Interchange Format.

4. ISO/IEC 19794-6:2005, Biometric Data Interchange Formats –

Part 6: Iris Image Data.

5. ‘‘Effect of severe image compression on iris recognition perfor-

mance’’, John Daugman, Cathryn Downing, May 2007, SC37

reference number 37N2125.

6. Position Paper in Support of Retaining the Iris Polar Image Data

Format, 0733007 v0.1

7. M1 Letter Ballot, Document number M1/07-0633

Iris Super-Resolution

YUNG-HUI LI

,MARIOS SAV VID ES

Language Technolog y Institute, Carnegie Mellon

University, Pittsbu rgh, USA

Department of Electrical and Computer Engineering,

Carnegie Mellon University, Pittsburgh, PA, USA

Synonyms

Iris image enhancement by Super-Resolution method;

Super-Resolution for iris

Definition

Super-Resolution is an image processing technique

which takes input of a single or multiple low-resolution

images and produces a single or multiple high-resolution

images. By Super-Resolution processing, the quality

of images can be enhanced and the follow-up stage of

image processing (e.g., segmentation, object recogni-

tion, object tracking, or biometric identiﬁcation) can

achieve a higher success rate. The goal of iris Super-

Resolution is to apply Super-Resolution technique in

the speciﬁc domain as in iris image in order to en-

hance the quality of iris image. The iris image of

better quality will result in a higher veriﬁcation/rec-

ognition rate in iris recognition systems.

Introduction

Image resolution is a fundamental factor for the suc-

cess of all kinds of image processing techniques, rang-

ing from

▶ image segmentation, ▶ object recognition,

tracking, 3D shape estimation, and reconstruction and

biometric recognition. The higher resolution the input

image has, the more accurate the output will be.

Figure 1 shows the two iris images (in polar coordi-

nate) which are taken from the same eye. The upper

one is taken with a long distance camera. Because

the distance between the eye and the camera is far,

the resolution and the quality of the image is not very

good. Therefore, many of the iris details cannot be

shown in this picture. On the other hand, the lower

image is taken with a high-resolution camera, placed

closer to the eye. Comparing the image quality among

these two pictures, one can clearly see that there are

much more details in the second picture. When these

two pictures are used as input to iris recognition sys-

tem, the lower one will give a more reasonable score.

Although the maximal resolution of digital cameras

has continually increased and the prices of them are

Iris Super-Resolution. Figure 1 Two iris images of the

same eye, shown in polar coordinate. Upper: iris image

taken from a long distance. Therefore, the image resolution

and quality is not very high. Lower: iris image taken from a

close camera, which contains much more details.

856

Iris Super-Resolution

continually decreasing, there are still times that the re-

gion of object of interest in a picture is very small, and

the resolution of the object is not high enough. In

such circumstances, in order to enhance the object of

interest and make the image of the object clearer,

image processing technique needs to be applied to

achieve this goal. Typically, in the ﬁeld of image pro-

cessing, those technique which takes input of one or

multiple low-resolution images and output one or mul-

tiple high-resolution images are called image Super-

Resolution.

Another problem that people have in their photos

very often is the blurring of the images. The blur of the

images might result from two reasons. One of them is

the relative motion between the camera and the object.

This is called motion blur. The other is the object is

out of focus of the camera during image acquisition

process. This is called focus blur. Besides image blur,

another problem we need to overcome in order to have

high quality image is the noise. Strictly speaking, the

problems of image blur and image noises are different

than the goal of Super-Resolution problem, however, in

the literature, in order for Super-Resolution algorithm

to be robust in real application, algorithm designers

need to take into account the problem of image blur

and image noises too.

Super-Resolution

Algorithms for Super-Resolution image can be categor-

ized into two different types. The ﬁrst is reconstruction-

based algorithms [1–5]. In general, this type of

Super-Resolution algorithm formulates the problem

of Super-Resolution as a matrix inversion problem.

Because digitiz ed images consist of pixels, a digitalized

image can be treated as a huge matrix, where each pixel

associates with a real number value in a limited range.

Therefore, a high-resolution image can be treated as a

matrix of large size, while a low-resolution image can

be treated as a matrix of small size.

A matrix can be converted to a vector, simply by

concatenating all the columns (or rows) into a long

vector. If we denote vector of the high-resolution

image as X and the low-resolution image as Y, Super-

Resolution problem can be simply formulated as

Y ¼ MX þV

where M is a transformation matrix that represents

the effect of imaging system and V is another column

vector of the same size of Y, which represents

the random noise inherent to any imaging system.

M can be further decomposed into several factors to

model different effects an imaging system produced

to the original high-resolution image; those effects

may include: relative motion between object and cam-

era, camera blur effect, color ﬁlter effect, and down-

sampling process.

The goal is to estimate the high-resolution image X

from observed low-resolution image Y. This goal can

be achieved by means of optimization. There are many

different goals that can be chosen to be optimized;

one of the most common ways is to minimize the

▶ L2 norm of the residual vector, which gives us the

following equation:

X ¼ arg min

Y  MX

Different algorithms give different ways and different

constraints to this optimization problem, and therefore,

result in different solutions. In summary, reconstruc-

tion-based Super-Resolution algorithms formulate the

problem as a matrix inversion problem and solve it by

various optimization methods.

Second type of Super-Resolution algorithms is prob-

ability-based algorithms [6–10]. In this type of algo-

rithm, the problem of reconstructing high-resolution

image from low-resolution ones is modeled by a Bayes-

ian approach. By Bayes’ rule, the probability of Super-

Resolution images given low-resolution ones Pr[Su|

] can be decomposed into:

Pr SujLo

½¼

Pr Lo

jSu½Pr Su½

Pr Lo

½

Since Pr[Lo

] is a constant and logarithm function is

a monotonically increasing function, the goal of Super-

Resolution can be achieved by this equation:

arg max

Pr½SujLo

¼arg max

ðln Pr½Lo

jSuln Pr½Su Þ

Different algorithms are trying to learn Pr[Lo

|Su]

from training data with different methods, and

integrated with different ways. But overall, the pro-

bability-based Super-Resolution algorithm is model-

ing the problem with a Bayesian approach and

solves it with the

▶ maximum a posteriori (MAP)

estimation.

Iris Super-Resolution

857

Iris Super-Resolution

Iris Super-Resolution is a very new topic in ﬁeld of

image processing and biometric recognition. There is

no relevant information found in the literature. In the

following sections, a rough idea of how to do iris

Super-Resolution is introduced to suggest a possible

way of enhancing iris image quality by fusing informa-

tion from multiple low-resolution images.

Information fusion can be thought of as a problem

in creating a high-resolution iris image from multiple

low-resolution images. Each one of the low-resolution

images has part of information that is needed, and the

goal is trying to extract those information needed and

put them at the right position on two-dimensional

space and ﬁnally create a pattern which gives an iris

image of higher resolution.

The process of painting can be used as a metaphor

to illustrate the process of iris Super-Resolution.

When a very high-resolution painting is to be painted

by ﬁrst looking at a few low-resolution images and ﬁll

in more local details, the ﬁrst thing is to select one

best image from the given low-resolution image. Let it

be called template image, and all other low-resolution

image are scene images. Secondly, the template image

is interpolated with zeros in order to enlarge to a

bigger size. The new size of the is a parameter which

can be ﬁne tuned later. After the process of enlarging

the image, zero-value pixel will spread evenly in entire

image. They are called holes.

After the bigger image with holes is created, next

step is trying to ﬁll up every hole with appropriate

numerical values, which are derived from combining

useful information from other low-resolution images.

In most cases, the appropriate numerical values for a

hole can be inferred from the regions that surround it,

and inside this region, the farther the points are away

from this hole, the smaller impact they have toward this

hole. Therefore, it is better to use a locality-based algo-

rithm to solve this problem. One way of processing

image locally is to cut the whole image plane into

smaller blocks. For example, if an iris image is of

size 30  360, and is cut into blocks of size 10  10,

there will be 3  36 ¼ 108 small blocks totally. Those

blocks are called ‘‘patches.’’ Processing image in patch-

based algorithms has been widely used in variety of

image processing ﬁeld, such as texture analysis, edge

and boundary detection, image segmentation, object

recognition, biometric recognition, and generic image

Super-Resolution.

The patch-based Super-Resolution algorithm is

quite intuitive. The ﬁrst step is to cut every scene

image into smaller patches. The second step is that

for every location, align the local patch of every scene

image with the template image. This step is important

because iris images usually suffer from image deforma-

tion problem. This is especially true for segmented iris

images since the pupil of an eye dilates when there is

strong ambient lighting, and contracts when ambient

lighting is weak. Therefore, it cannot be naively as-

sumed that every patch from different images corre-

sponds to exactly the same position on iris surface.

Patches from different images have to be aligned, so

that every pixel on the scene patch can be mapped

correctly to the template image in the corresponding

location.

After patches are locally aligned with each other,

the third step is to combine information about nu-

merical value for each pixel from scene patches and

ﬁll the holes on template patches with new value. If

this process is illustrated with the metaphor of paint-

ing, this step is to draw details at the blank region on

the template image. There are dozens of different

algorithm to fuse information. One of the easiest

ways is the method of linear combination. Suppose

there are n numerical values from n different

scene patches to ﬁll one blank hole, the process of

linear combination can be expressed as the following

equation:

Y ¼ a

þ a

þþa

k¼1

where x

is the numerical value of the ith scene

patch, a

is the coefﬁcients of linear combination, and

Y is the numerical value of the pixel after linear

combination.

After all the blank holes are ﬁlled with the new

values, the last step is to perform image smoothing

across the boundaries of the patches. There are many

existing smoothing algorithms. Linear, quadratic, or

cubic interpolation can be used to achieve this goal.

After the interpolation, the high-resolution image can

be used or it can be down-sampled to make the ﬁnal

image the same size as the scene image but with much

more detailed texture within. Figure 2 illustrates the

ﬂow chart of iris Super-Resolution algorithm.

858

Iris Super-Resolution

Summary

Image Super-Resolution is an important topic in image

processing and has been extensively used in many app-

lications, including quality enhancement for images

and video, video surveillance, and biometric recogni-

tion. Super-Resolution for iris image is a very new topic.

The technique can be applied to both iris synthesis and

iris recognition. Not many detailed research results have

been reported. In this section, one iris Super-Resolution

algorithm is simply introduced by pixel-level informa-

tion fusion from multiple low-resolution images. By

performing Super-Resolution on iris image, the recog-

nition rate of iris recognition system can be improved,

and therefore, making the biometric system even more

accurate and suitable for many real-life situations where

high security issue is the top priority.

Related Entries

▶ Iris Image Quality

▶ Iris on the Move

▶ Iris Recognition, Overview

▶ Iris Sample Synthesis

References

1. Elad, M., Feuer, A.: Restoration of a single super-

resolution image form several blurred, noisy, and under-sam-

pled measured images. IEEE Trans. Image Process. 6(12),

1626–1658 (1997)

2. Irani, M., Peleg, S.: Improving resolution by image registration.

CVGIP: Graphics Models and Image Proc. 53, 231–239 (1991)

3. Schulz, R.R., Stevenson, R.L.: Extraction of high resolution

frames from video sequences. IEEE Trans. Image Process. 5,

996–1011 (1996)

4. Hardie, R.C., Barnard, K.J., Armstrong, E.E.: Joint MAP registra-

tion and high resolution image estimation using a sequence of

under-sampled images. IEEE Trans. Image Process. 6, 1621–1633

(1997)

5. Farsiu, S., Robinson, D., Elad, M., Milanfar, P.: Advances

and challenges in Super-Resolution. Int. J. Imaging Syst. Tech-

nol. 14(2), 47–57 (2004)

6. Baker, S., Kanade, T.: Limits on super-resolution and how to

break them. IEEE Trans. Pattern Anal. Mach. Intell. 24,

1167–1183 (2000)

7. Capel, D.P., Zisserman, A.: Super-resolution from multiple views

using learnt image models. Proceedings of IEEE International

Conference, Computer Vision and Pattern Recognition 2,

627–634 (2001)

8. Liu, C., Shum, H., Zhang, C.: A two step approach to hallucinat-

ing faces: global parametric model and local nonparametric

model. Proceedings of IEEE International Conference on Com-

puter Vision and Pattern Recognition, 192–198 (2001)

9. Freeman, W., Pasztor, E.: Learning low level vision. In: Proceed-

ings of the Seventh International Conference on Computer

Vision, 1182–1189 (1999)

10. Baker, S., Kanade, T.: Hallucinating faces. Proceedings of the

International Conference on Automatic Face and Gesture Rec-

ognition, 83–88 (2000)

Iris Template Extraction Via Bit

Inconsistency and GRIT

GERRY VERNON DOZIER

,MARIOS SAVVIDES

KELVIN BRYA N T

,TAIHEI MUNEMOTO

KARL RICANEK,JR.

,DAMON WOODARD

North Carolina A&T State University

Carnegie Mellon Universit y

University of North Carolina at Wilmington

Clemson University

Synonyms

Bit fragility; Bit inconsistency; Fragile bits

Definition

The charact eristic of iris code bits values being incon-

sistent (also referred to as fragile) across different

images of the same iris was explored by Hollingsworth

et al. [1]. The notion of fragile bits was ﬁrst suggested

Iris Super-Resolution. Figure 2 The flow chart of the iris Super-Resolution algorithm.

Iris Template Extraction Via Bit Inconsistency and GRIT

859