IRIS IMAGE SEGMENTATION BASED

ON INDUSTRIAL VISION TOOLS

Silvia Anton and Florin Daniel Anton

Department of Automation and Industrial Informatics, University Politehnica of Bucharesty

Spl.Independentei, 313, Bucharest, Romania

Keywords: Iris recognition system, Image processing, Iris image segmentation, Intelligent vision tools, Rapid

deployment applications.

Abstract: In the last few years biometric data acquisition and processing systems for person identity verification and /

or identification started to be increasingly used. This is done both in military applications for person

identification in military operations and war theatres, but also in civilian applications for personal identity

verification, accounting systems, etc. Depending on the organization policy, such systems must be secured

and customized, for application enhancement, and to fulfil the organization requirements. Such systems

which allow customization and enhancement are not available for source code modification and feature

enhancement. This paper presents a software image processing development environment IPDE based on

vision tools, which is able to run vision projects but also allow the user to develop stand alone applications

in a short amount of time, applications which are based on customized vision tools. The IPDE is used to

exemplify the process of creating an iris recognition application where a set of vision tools were used in

order develop a customized iris image segmentation routine. The paper is structured on three chapters

presenting the IPDE architecture, the vision tools, the application development stages, and ends with some

experimental data and conclusions.

1 INTRODUCTION

In the last years, the identification and verification of

the identity of persons has become an increasingly

important factor (Zhai, 2009; Patnala, 2009; Araghi,

2010; Matschitsch, 2008). A special place is held by

systems for person identity verification /

identification based on the iris, which were accepted

and used especially in military applications (L-1

Identity Solutions Inc., 2010).

The biggest advantage of using the iris as a

biometric verification and recognition method is the

accuracy and reliability (Daugman, 2004) estimated

to be ten times more accurate than methods using

fingerprint, iris-based methods produce a false

match rate (or false acceptance rate – FAR) of 1/1-2

million samples, while fingerprint-based methods

produce a false match rate close to 1/100000

samples (Cao, 2005; Ganeshan, 2006;). Due to this

aspect some countries have initiated the procedures

to integrate the iris biometric data into the

population identification cards.

While fingerprints are constantly exposed and

are likely to deteriorate, the iris is naturally protected

by the cornea (a transparent membrane covering the

eye) and its model seems to remain unchanged for

decades, being only affected by some eye diseases

which are more frequently found to elders,

population which is less probable to be involved in

such identification process.

Unlike fingerprint scanners, which require direct

contact and must be kept extremely clean, iris scan

can perform safely and hygiene at some distance

from the eye. Disadvantages include the iris

scanning higher initial cost (few thousands of

dollars) and the fact that it is still a relatively new

technology that has not been tested enough.

Also to implement this technology some

organizations require special software products

developed “in the house”. In order to do this in the

shortest time, an IPDE which allow rapid application

development is needed.

This paper gives some solutions to the issues

presented and offers a rapid implementing solution

for acquiring and processing iris information. In the

333

Anton S. and Daniel Anton F..

IRIS IMAGE SEGMENTATION BASED ON INDUSTRIAL VISION TOOLS.

DOI: 10.5220/0003572803330336

In Proceedings of the 8th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2011), pages 333-336

ISBN: 978-989-8425-75-1

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

next sections the software architecture of the

proposed solution is described, followed by the

presentation of the vision tools used and examples.

2 THE SOFTWARE

ARCHITECTURE

In order to rapidly develop an iris recognition

application which can be customized by the

organization, the image processing subsystem must

be accessible for modifications. Only few IPDE

have this property, and have been considered for

implementation, in the end the chosen system was

AdeptSight from multiple reasons:

• The vision system have high performances in

object localisation: 1/40 of a pixel in position,

and 0.01 degree in rotation (which allow a

precise iris, pupil and eyelid localization)

• The system has the possibility to train and edit

object models based on non connected contours

(this is the case of the iris, which in most

situations is partially occluded by the eyelid)

• The programming interface is based on visual

tools (visual programming) for rapid

development, and has also the possibility to be

integrated with high level programming

languages (C#) for complex applications.

• The system can be easily integrated with other

systems by using Ethernet, serial, or I/O lines.

(Fig. 1)

Figure 1: External device - Vision System integration.

AdeptSight uses up to four FireWire cameras

connected directly to the PC where the AdeptSight

software is installed.

The development of vision applications is based

on vision projects which have the following

structure: The vision project is separated in two

main parts, the first part handles the hardware and

the communication environment and is composed by

communication routines and configuration of system

devices like cameras (Basler, Direct Show or

Emulation – virtual camera), and other hardware

devices. The second part is represented by the vision

sequences, which compose the principal part of the

project. The sequences are composed by vision tools

connected together and which are executed step by

step in a sequence defined by the programmer. In

addition the user can develop C# programs which

interact with the AdeptSight project and extend his

functions regarding the communication with other

devices and other functions which are not

implemented in AdeptSight.

3 VISION PROJECT

IMPLEMENTATION

The vision is calibrated using a dotted pattern which

is placed in front of the camera at a distance which

approximates the distance to the subject eye. The

calibration is executed using a 2D camera calibration

wizard which guides the user step by step through

the calibration process. The single information

which the user must supply is the Dot Pitch of the

calibration pattern, the rest of the process is handled

by the wizard. After the 2D camera calibration the

following information’s are obtained:

• Average Pixel Width/Height

• The lens distortions are corrected

• The perspective distortion is also corrected

After the 2D camera calibration, the sequence

can be loaded into the project, and for this

application the sequence is very simple and consists

on the following vision tools, an acquisition tool

which obtain the image from the camera, a set of

two localization tools named Locator(s) which have

the role to recognize the pupil and the iris

boundaries, and a set of two locators which will

detect the eyelid boundaries, which are required in

order to obtain only the iris valid image date.

In this application the difficulty is to detect the

correct boundaries of the iris disregarding the outer

irregularities which are generated by the contrast

between the iris and sclerotic membrane and the

eyelids.

Fig. 2 presents the interface for model editing.

The system, based on the contrast threshold, and the

outline and detail levels detect the contours and

proposes the user a set of contours for model

building. The outline level provides a coarser level

of contours than the detail level. The location

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

334

Figure 2: Editing the model for the lower eyelid.

process uses outline level contours to rapidly

identify and roughly locate potential instances of the

object, then, the location process uses the detail level

contours to confirm the identification of an object

instance and refine its location within image.

The user can modify the selected contours by

deleting/adding contours, or select only parts of the

proposed contours (Adept, 2001, Adept, 2007). In

Fig. 2 the red line is selected for deletion, the blue

contours are marked as deleted (will not be used for

model building) and the green contours are valid

contours. After all contours have been selected, the

model can be build and used for recognition.

4 CONCLUSIONS

The proposed IPDE is a software development

environment for image processing which allow rapid

application development and in the same time allows

the developer to customize the vision tools which

the IPDE offers by default.

The AdeptSight IPDE is a fast solution for

developing “in house” customized iris recognition

applications for organizations which require a high

level of personal identification system customization

and optimization. The IPDE offers a set of default

vision tools which can be customized by changing

the parameters in the vision project but also allows

the developer to write applications in which those

tools can be used and moreover modified to satisfy

the application necessities.

In our case the application demonstrates the

rapidity of image processing of the vision tools (all 5

tools – 1 acquisition tool, and 4 locators, have been

executed in less than 30 milliseconds). Also we

demonstrated the reliability of the vision tools 98%

of the images which had the iris in the centre of the

image have been correctly segmented (the pupil, iris,

and eyelids have been correctly and accurately

detected) (the tests have been conducted on Bath Iris

image database) (University of Bath, 2009). In the

rest of the images where the iris was not centred the

system have problems in detecting the eyelids due to

the angle which the axis of the camera makes with

the eye, the eyelids borders being masked by the hair

on the eyelid.

Figure 3 presents some results in detecting the

pupil, iris and eyelids basic features in iris

segmentation. The used method can be as good as

IRIS IMAGE SEGMENTATION BASED ON INDUSTRIAL VISION TOOLS

335

Figure 3: Results in detecting the pupil, iris, and eyelids for iris image segmentation.

other classic segmentation methods (sobel filters,

Gabor and long Gabor filtering) (Daugman, 2007;

Popescu-Bodorin, 2010).

ACKNOWLEDGEMENTS

The work has been co-funded by the Sectoral

Operational Programme Human Resources

Development 2007-2013 of the Romanian Ministry

of Labour, Family and Social Protection through the

Financial Agreement POSDRU/89/1.5/S/62557.

REFERENCES

Adept Technology Inc., 2001, AdeptVision Reference

Guide Version 14.0 Part Number 00964-03000, San

Jose, Technical Publications.

Adept Technology Inc. 2007, AdeptSight 2.0 Online Help,

San Jose, California, USA.

Araghi, L. F., Shahhosseini, H., Setoudeh, F., 2010. Iris

Recognition Using Neural Network. Proceedings of

The International MultiConference of Engineers and

Computer Scientists 2010 Vol I, IMECS 2010, March

17-19, Hong Kong

Cao, W., Hu, J., Xiao, G., Wang, S., 2005. Iris

Recognition Algorithm Based on Point Covering of

High-Dimensional Space and Neural Network,

MLDM 2005, LNAI 3587, pp. 305 – 313, Springer-

Verlag Berlin Heidelberg 2005

Daugman, J. G., 2004. How iris recognition works, IEEE

Trans. On circuits and Systems for Video Technology,

vol. 14, no. 1, pp. 21-30.

Daugman, J. G., 2007. New methods in iris recognition,

IEEE Trans. On Systems, Man, and Cybernetics - part

B: Cybernetics, vol. 37, no. 5., pp. 1167-1175.

Ganeshan, B., Theckedath, D., Young, R., Chatwin, C.,

2006. Biometric Iris Recognition System Using A Fast

And Robust Iris Localization And Alignment

Procedure, Optics and Lasers in Engineering 44

(2006) 1–24

L-1 Identity Solutions Inc., 2010. HIIDE 5 Solutions,

http://www.l1id.com/pages/774-hiide-5.

Matschitsch, S., Stogner, H., Tschinder, M., 2008.

Rotation-Invariant Iris Recognition Boosting 1d

Spatial-Domain Signatures To 2D, ICINCO 2008 -

International Conference on Informatics in Control,

Automation and Robotics.

Patnala, S. R., Murty, C., Reddy, E. S., Babu, I. R., 2009.

Iris Recognition System Using Fractal Dimensions of

Haar Patterns. International Journal of Signal Pro-

cessing, Image Processing and Pattern Recognition,

Vol. 2, No.3.

Popescu-Bodorin N., Balas, V. E., 2010. Comparing Haar-

Hilbert and Log-Gabor Based Iris Encoders on Bath

Iris Image Database. IEEE SOFA 2010. (http://fmi.

spiruharet.ro/bodorin/articles/ieee-sofa-2010-bodorin-

balas.pdf)

University of Bath Iris Database. (2009) http://www.bath.

ac.uk/eleceng/researc

ICINCO 2011 - 8th International Conference on Informatics in Control, Automation and Robotics

336