Apresentação - Iris Recognition

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 1/37

IRIS RECOGNITION

Igor Leonardo O. Bastos –  igorcrexito@gmail.com

Handbook of Biometrics –  Chapter 4

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 2/37

Summary

 Understanding an eye

 A short history about iris recognition

 Current state

 The Method

Off-Axis Gaze Correction

Detecting and Excluding Eyelashes

Evaluation

 References

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 3/37

UNDERSTANDING

AN EYE

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 4/37

Understanding an eye

 How is called any part of an eye?

Fig 1 –  Eye parts and its names

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 5/37

A SHORT HISTORY OF

IRIS RECOGNITION

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 6/37

A Short History of Iris Recognition

 Iris was the target of studies since the ancient Egypt,

Chaldea and ancient Greece

Stone inscriptions, painted ceramic artifacts, writings

 Commonly associated to the art of ‘Divination’

Fig 2 –  Symbol of protection and royal power

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 7/37

A Short History of Iris Recognition

 Studies about iris remits to Hippocrates writings and

Imothep’s

Fig 3 –  Hippocrates and Imhotep

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 8/37

A Short History of Iris Recognition

 Alphonse Bertillon iris as a distinguishing human

identifier

 James Doggartcomplexity of iris patterns and

adequacy to serve in the same way of fingerprints(oneness)

 Flom and Safir patent of Doggart’s concept but no

method to make it possible

Fig 4 –  Alphonse Bertillon, Leonard Flom and Aran Safir

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 9/37

CURRENT STATE

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 10/37

Current State

 Developed rapidly over the past 15 years Large number of active researchers in academy and

industry

 Lots of people enrolled in iris recognition systems

 Systems are usually designed for use in

identification-modeOne-to-many exhaustive search

Astonishing number of comparisons

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 11/37

Current State

 Basic research into alternative methods continues

 Scientific and engineering literature about iris

recognition grows monthly

Contributions from several university and industrial labs

around the world

 Iris recognition systems employed by government

agencies

Project IRIS from UK to identify immigrants

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 12/37

Current State

 Many iris recognition datasets are available

CASIA Iris Image Database (v4)

UPOL Iris Database

UBIRIS Database

ICE 2005 Database

BATH University Iris Database

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 13/37

THE METHOD

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 14/37

The Method

 Finding the Iris is the first thing to be done

Inner and outer boundaries at pupil and sclera

Upper and lower eyelids

Superimposed eyelashes or reflections from the cornea oreyeglasses

 Precision in assigning the true inner and outer

boundaries is a critical operation Innacuracy can cause different mappings of the iris pattern

in its extracted description

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 15/37

The Method

 Important point iris is not an annulus

Inner and outer boundaries are usually not concentric

 Pupil boundaries and outer iris boundaries are non-circular

 Performance improved by relaxing bothassumptions

Methods for modelling boundaries whatever their shapes

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 16/37

The Method

 Finding the correct countours can be difficult

Eyelids can occlude iris outer boundary

Reflections from illumination can occlude iris inner boundary

Both can be occluded by reflections from eyeglasses

 It is necessary to fit flexible contours that can

tolerate interruptions

 A constraint both boundaries are closed curves

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 17/37

The Method

 Employment of Active Contours based on discrete

Fourier series expansions of the contour data

 Selecting the number of frequency componentsallows control over the degree of smoothness that is

imposed

Truncating means applying a low-pass filtering over theboundary curvature data

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 18/37

The Method

Fig 5 –  Snakes and corresponding maps

 Snakes representing pupil and iris boundaries

Ideal snakes would be flat and straight

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 19/37

The Method

 There is a tradeoff between simplicity of the model

and its precision

Number of terms used on the Fourier Series (M)

 Empirically, the author found good approximations

for values M=17 (pupil) and M=5 (iris)

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 20/37

OFF-AXIS GAZE

CORRECTION

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 21/37

Off-Axis Gaze Correction

 Model requires an on-axis image of the eye

Stop-and-stare interface

 Correcting the projective deformation on the iriswhen its image is off-axis

 The essence of the problem is estimating the anglesof gaze relative to the camera

Eye morphology is so variable that is unlikely that this factor

could support a robust estimation

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 22/37

Off-Axis Gaze Correction

 Obvious alternative approach shape of pupil

 Estimating the ellipticity of the pupil and orientation

of that ellipse would yield the gaze deviation

 The author proposes something different Fourier-

based trigonometry

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 23/37

Off-Axis Gaze Correction

 Fourier series expansions of the X- and Y-

coordinates of the pupil contains shape distortion

information related to the deviated gaze

 If the pupil boundary is a circle, this method is

reduced to the previous one

 Estimating these parameters gaze deviation to

be reversed by an affine transformation of the off-

axis image

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 24/37

Off-Axis Gaze Correction

Fig 6 –  Off-Axis Gaze Correction

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 25/37

DETECTING AND

EXCLUDING EYELASHES

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 26/37

Detecting and excluding eyelashes

 Eyelashes can occlude parts of the eye

  Usually have random and complex shapes

 Can be the strongest signals in iris images, in terms

of contrast and energy

  They could dominate the image with spurious information

Fi 7 –  E elashes occludin arts of the e e

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 27/37

Detecting and excluding eyelashes

 Employment of statistical estimation methods that

depend essentially on determining whether the

distribution of iris pixels is multi-modal

 If the lower tail of the iris pixel histogram supports

na hypothesis of multi-modal mixing, then an

appropriate threshold can be computed

Fi 8 –  Histo ram and threshold com utation

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 28/37

EVALUATION

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 29/37

Evaluation

 Comparison between IrisCodes (bit pair Bernoulli

trials)

 Areas obscured by eyelids, eyelashes or byreflections from eyeglasses are processed and

prevented to influence the iris comparisons

 IrisCodes keep the information about the phases

and are compared through bit-wise mask functions

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 30/37

Evaluation

 Code related to each iris is ExclusiveOR’ed and

AND’ed to mask functions

 Raw Hamming distance used to compared to irises:

 The number of bits pairings available for

comparison is usually nearly a thousand

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 31/37

Evaluation

 Restrictions related to the ‘behaviour’ of masks

 Must take into account the amount of comparison

data that was available

 A normalization is performed in order to improve

the confidence level score normalization

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 32/37

LARGE-SCALE

APPLICATIONS

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 33/37

Large-Scale applications

 Score distribution for 200 Billion Different Iris Comparisons

Fig 9 –  Hamming Distance of different irises

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 34/37

Large-Scale applications

 Use of thresholds to compute the similarity of one iris to

another

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 35/37

OBRIGADO !

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 36/37

REFERENCES

8/17/2019 Apresentação - Iris Recognition

http://slidepdf.com/reader/full/apresentacao-iris-recognition 37/37

References

1. John Daugman. Iris Recognition. In: Handbook of Biometrics,

Springer, USA (2008)

2. Enrique A. Velasco. Connections in Mathematical Analysis: The

Case of Fourier Series. In: American Mathematical Monthly,v.99, USA (1992).

3. Michael Kass, Andrew Witkin and Demetri Terzopoulos.

Snakes: Active Countor Models. In: International Journal ofComputer Vision, p. 321-331. (1998)