Diapositive 1

1
Person Identification technique using human iris
recognition
Christel-loïc TISSE1, Lionel Martin1, Lionel
TORRES2, Michel Robert2
1 Advanced System Technology STMicroelectronics 2
Université de Montpellier , Lirmm
Presented By BOLTZ Sylvain
2
Introduction Iris Biometrics

• Iris great for authentification probability of
  two same iris pattern is almost 0 , iris
  protected from environment , aging
• Locating iris often fails even on cooperative
  person
• Local features extraction Daugmans system not
  fully publicated in litterature, new approach on
  recent 2D-Hilbert transform works.

3
Locating Iris Daugmans System

• Use of a circular edge detector
• Daugmans Localization Sytem appears to succeed
  only with 86 ratio.

Intro-Differential Operator
The main fail cause is Spot reflection in the
pupil.
4
Locating Iris Circular Hough Transform

• Detection strategy combination of the
  integro-differential operator and a Hough
  transform
• First Hough transform to locate the center then
  integro-differential operators
• Circular Hough transform

5
Cartesian to polar reference transform

• We unfold the circular picture into a
  dimensionless rectangular image

• Pupil not perfectly circular
• Iris outer boundary can be faked by contact lens.

6
Local feature extraction Prelude

• Daugmans System Gabors complex 2D Filters
• Iris code emergent frequency, instantaneous
  phase.
• Problem They are not defined for a real signal
• Solution Analytic Image. (fast algorithm)
• Where H is Hilbert Transform

7
Local feature extraction
Rectangular Iris Image NxM
Real valued Signal The iris image.
8
Local feature extraction
Rectangular Iris Image NxM
2D Pass-Band Filter

• 2D Pass-Band filter must not dephase the signal.
• We use a 2D Hamming window
• X(u,v)X1(u)X2(v)
• Where X1(u) and X2(v) are Well-Known
• 1D Hamming window.

9
Local feature extraction
Rectangular Iris Image NxM
2D Pass-Band Filter

• We build the analytic image of the output Signal

10
Local feature extraction
Emergent Frequency
Rectangular Iris Image NxM
2D Pass-Band Filter
11
Local feature extraction
Rectangular Iris Image NxM
2D Pass-Band Filter
12
Local feature extraction
Rectangular Iris Image NxM
2D Pass-Band Filter
Same on 3 different Hamming Filters
13
Local feature extraction
Rectangular Iris Image NxM
2D Pass-Band Filter
Iris Code
Final Output Binary Images NxM by
Thresholding. Is the IRIS CODE. Then a simple
Hamming distance test between the template and
the extracted code.
14
Experimental results
15
Experimental results

• Iris Localisation
• On a 300 images various iris database (contact
  lens , glasses, )
• Without Hough transform preprocessing 86
• With Hough transform preprocessing 100
• Complexity

16
Experimental results

• False Acceptance Rate and False Reject Rate
• this estimation of FAR and FFR is without
  enrolment
• FAR0 gt FRRlt3( 8 bad localization)

17
Conclusion, Possible Improvments and Critics.

• Same Performance as Daugmans System in local
  feature extraction
• Great amelioration on Locating iris with Circular
  Hough Transform.
• Too small test database
• In progress
• memory cost