Iris Identification Using Wavelet Packets - Identification des

Document Sample
Iris Identification Using Wavelet Packets - Identification des Powered By Docstoc
					        Iris Identification Using
            Wavelet Packets

Emine Krichen, Mohamed Anouar Mellakh, Sonia
      Garcia Salicetti, Bernadette Dorizzi

{emine.krichen,anouar-mellakh;sonia.salicetti;bernadette.dorizzi}@int-
                              evry.fr

          Institut National des Télécommunications
         9 Rue Charles Fourier , 91011 Evry France



                                                                         1
              Outline
• Classical approach versus our
  approach (Packets Method)
• Experimentations on 2 databases
• Introduction of color information
• Conclusion and perspectives



                                  2
            Introduction
• Study of iris recognition on normal
  light illumination
   –Use of usual devices
   –Fusion between iris and other
     biometric modalities (face, eye
     shape…)


                                        3
Comparison infra-red / normal light




    Normal light           Near Infra red
• Lack of texture information
• Presence of a great number of reflections 4
                 Iris Segmentation




Circular Edge detector     Hough Transform (Iris circle)   5
                                 Wavelet method
   • 2D wavelet basis : Gabor
        iωθ0 φ  r0 ρ 2 α2 θ0 φ 2 β2
  e             e             e                 Iρ, φρ dρ dφ
  

   • Spatial parameters in
     polar coordinates (ρ,θ).
   • 4 resolution levels
   • 2048 coefficients for
     coding the iris.


       J. Daugman, “How iris recognition works”, Proceedings of the International   6
Conference on Image Processing, 22-25 September 2002
      Our approach : Packet method
• Process the whole
  image at each level
  of resolution
• Starting with higher
  mother wavelet
  window
• 1664 coefficients for
  coding iris

                                     7
            Databases
• IrisINT : Iris images recorded under
  normal       light illumination.  70
  persons 700 images.



• CASIA : Iris images taken under
  infra red illumination. 110 persons,
  770 images. Recorded at NLPR
  China.
                                         8
       Roc curves (IrisINT)




•Poor results for the wavelet method
•The wavelet Packet method is more
robust using visible light images      9
  Comparative results on CASIA
          and IrisINT
Databases                        IrisINT        CASIA
Type of errors             FAR       FRR     FAR     FRR
Classical wavelet method   2%       12.04%   0.35%   2.08%
Packets method             0%        0.57%   0.2%    1.38%

• With infra red illumination, the two
methods     have     quite    the same
performance. WP is more robust to the
presence of eyelids or eyelashes.
                                                        10
                Use of color information
                            ACR method


  Original color image                                      Color image (256 colors)
(71.000 different colors)


 We perform iris recognition using
 the same algorithm as the one
 developed for grey level image
                            C.P. Strouthopoulos, Adaptive                      11
                                    color reduction
     Use of color information :
      ROC curve on IrisINT




Use of color information allows a better
discrimination between the persons.        12
 Conclusion and perspectives
• The packets method allows better
  performance on normal light
  illumination images.
• Color information can be used to
  improve results on simple grey
  level images.
• Results need to be confirmed using
  larger bimodal database (in order to
  decrease the variance).
                                    13
      Adaptive color reduction (ACR)




RGB +                                                                          One
neighborhood                                                                   Neuron
information                                                                    per color




                 Self organized neural network
         Reduction adapted to initial distribution of colors

N. Papamarkos, A.E. Atsalakis, and C.P. Strouthopoulos, Adaptive colour reduction, IEEE
   Transactions on Systems, Man, and Cybernetics, Vol. 32, N°1, , February 2002.     14