Iris Recognition - Amir Omidvarnia by zhouwenjuan


									Iris Recognition

      Amir Hosein Omidvarnia
             August 2006
• Introduction to Biometrics
• Anatomy Of Human Eye
• Iris Recognition System
• Iris Image Processing
        - iris localization
        - Feature Extraction
        - Matching
• Current Works
• References
 2 Categories of Biometrics
   Physiological – also known as static biometrics:
    Biometrics based on data derived from the
    measurement of a part of a person’s anatomy. For
    example, fingerprints and iris patterns, as well as facial
    features, hand geometry and retinal blood vessels
   Behavioral – biometrics based on data derived from
    measurement of an action performed by a person and,
    distinctively, incorporating time as a metric, that is, the
    measured action.         For example, voice (speaker
Biometrics – How do they work?
 Although biometric technologies differ, they all
  work in a similar fashion:
   The user submits a sample that is an identifiable,
    unprocessed image or recording of the physiological
    or behavioral biometric via an acquisition device (for
    example, a scanner or camera)
   This biometric is then processed to extract
    information about distinctive features to create a
    trial template or verification template
   Templates are large number sequences. The trial
    template is the user’s “password.”
Strengths, Weaknesses and
Usability of Biometrics
 Biometric     Strengths                    Weakness                        Usability

 Iris             Very stable over time       Potential user resistance      Information
                  Uniqueness                  Requires user training          security access
                                               Dependant on a single           control, especially
                                                vendor’s technology             for
                                                                            Federal Institutions and
                                                                               Physical access
                                                                                control (FIs and
                                                                               Kiosks (ATMs and
                                                                                airline tickets)
 Fingerprint      Most mature biometric       Physical contact                IS access control
                   technology                   required (a problem in          Physical access
                  Accepted reliability         some cultures)                   control
                  Many vendors                Hampered by temporary           Automotive
                  Small template (less         physical injury
                   than 500 bytes)
                  Small sensors that can
                   be built into mice,
                   keyboards or portable
Strengths, Weaknesses and
Usability of Biometrics
 Biometrics   Strengths                     Weakness                       Usability

 Voice           Good user acceptance         Unstable over time            Mobile phones
                 Low training                 Changes with time,            Telephone banking
                 Microphone can be             illness stress or injury       and other
                  built into PC or mobile      Different microphones          automated call
                  device                        generate different             centers
                                               Large template
                                                unsuitable for

 Signatures      High user acceptance         Unstable over time            Portable devices
                 Minimal training             Occasional erratic             with stylus input
                                                variability                   Applications where
                                               Changes with illness,          a “wet signature”
                                                stress or injury               ordinarily would be
                                               Enrollment takes times         used.
Strengths, Weaknesses and
Usability of Biometrics
Biometrics   Strengths                   Weakness                        Usability

Face            Universally present        Cannot distinguish             Physical access
                                             identical siblings              control
                                            Religious or cultural
Hand            Small template             Physical size of               Physical access
                 (approximately 10           acquisition device              control
                 bytes)                     Physical contact required      Time and
                Low failure to enroll      Juvenile finger growth          attendance
                 rate                       Hampered by temporary
                Unaffected by skin          physical injury
Retina          Stable over time           Requires user training         IS access control,
                Uniqueness                  and cooperation                 especially for high
                                            High user resistance            security government
                                            Slow read time                  agencies
                                            Dependent on a single          Physical access
                                             vendor’s technology             control (same as IS
                                                                             access control)
Future Trends in Biometrics
 Body Odor – Body odor can be digitally recorded
  for identification. A British company, Mastiff
  Electronic System Ltd. Is working on such a system
 DNA Matching – The is the ultimate biometric
  technology that can produce proof positive
  identification of an individual
 Keystroke Dynamics – Keystroke dynamics, also
  referred to as typing rhythms, is an innovative
  biometric technology
Iris Recognition
Anatomy of the Human Eye

                 • Eye = Camera

                 • Cornea bends, refracts, and
                 focuses light.

                 • Retina = Film for image
                 projection (converts image into
                 electrical signals).

                 • Optical nerve transmits
                 signals to the brain.
Anatomy of the Human Eye

            The structure of the a transverse
  The structure of the iris seen inhuman eye section
   The structure of the iris seen in a frontal section
Individuality of Iris

    Left and right eye irises have distinctive pattern.
       How does Human Vision
           System work?
     A Top to down scenario                A Bottom-up scenario
     I see a human body

                                          components (such as
  I expect to see a human face            edges, lines etc.)

I expect to see two eyes and a nose            objects

                     Two possible hypotheses
Iris recognition
 Processing flow
         Acquire                           Analyze
         Image                           Image Data
                          Verification                    Enrollment
         Define                Present                 Store
   Limbus-Iris Boundary   Identification code     Reference code

           Define                       Compare
    Pupil-Iris Boundary            by Hamming Distance

        Establish                    Calculate Decision
    Coordinate System                Confidence Level

         Define                           Accept or
      Analysis Band                        Reject
Analysis & Recognition
                      Analysis &

  Image         Image            Feature
 Selection   Enhancement        Extraction
Analysis & Recognition
                         Analysis &

  Image            Image            Feature
 Selection      Enhancement        Extraction

 We assume that several iris images have been captured
 Must remove blurred / occluded images before feature
 Automatic methods exist
Analysis & Recognition
                            Analysis &

     Image           Image             Feature
    Selection     Enhancement         Extraction

   Features in initial image are very subtle and difficult to
   Typically, operations are applied to increase contrast,
Analysis & Recognition
                      Analysis &

  Image         Image            Feature
 Selection   Enhancement        Extraction
Analysis & Recognition
                          Analysis &

     Image          Image           Feature
    Selection    Enhancement       Extraction

   Use wavelet transform and etc to locate sharp variations
    in local intensity levels
   As an example, Collect the locations of sharp variations
    in a “feature vector”
Analysis & Recognition
                          Analysis &

     Image          Image            Feature
    Selection    Enhancement        Extraction

   Convert feature vector to a binary representation
   Compare to other iris images in database to find a match
    and identify the person
Analysis & Recognition

            Acquisition          Image

 IrisCode   Gabor Filters   Polar Representation
                                                   Demarcated Zones
Iris Imaging

               • Distance up to 1 meter

               • Near-infrared camera
Image Acquisition

 Accurate setup of equipment and eye positioning
 Quite bulky.
 Up to 1m possible with video cam + telephoto lens.
Iris Image Processing
         Edge Detection

• Why detect edge?
   Edges characterize object boundaries and are
   useful features for segmentation, registration
   and object identification in scenes.

• What is edge?
          No rigorous definition exists

Intuitively, edge corresponds to singularities in the image
(i.e. where pixel value experiences abrupt change)
        Gradient Operators

 • Motivation: detect changes

   change in the pixel value        large gradient

image       Gradient                                  edge
            operator                                  map
 x(m,n)                g(m,n)                        I(m,n)

          MATLAB function: > help edge
     Common Operators

• Gradient operator

Examples: 1. Roberts operator

             g1                 g2
Common Operators (cont’d)

  2. Prewitt operator   3. Sobel operator


Compass Operators
      Laplacian of Gaussian

 • Generalized Laplacian operator

                Gaussian             Laplacian            edge
                LPF ()              operator             map
 x(m,n)                     g(m,n)                       I(m,n)

Pre-filtering: attenuate the noise sensitivity of the Laplacian
            Zero Crossings

                                             zero crossing

        f            f’                    f’’

image       Laplacian                             edge
            operator                              map
 x(m,n)              g(m,n)                      I(m,n)
Robust Laplacian-based Edge

                       estimate    2
                    local variance

        Laplacian           zero                     yes edge
image                                     2>th          point
        operator         crossing?
                                          not an
                           not an
                                        edge point
                         edge point
   Canny Edge Detector

                 Original image

    Smoothing by Gaussian convolution

 Differential operators along x and y axis

        Non-maximum suppression
      finds peaks in the image gradient

Hysteresis thresholding locates edge strings

                 Edge map
Hough Transform
    For any (x, y) there is a one parameter family of lines through
     this point, given by

    Each point gets to vote for each line in the family; if there is a
     line that has lots of votes, that should be the line passing
     through the points
     Hough Transform

tokens           votes
Hough Transform
 A line is the set of points (x, y) such that

             tokens                         votes
Hough Transform
Iris Localization
I. Daugman
   The Daugman system fits the circular contours via
    gradient ascent on the parameters so as to

    Where                                   is a radial Gaussian,
    and circular contours (for the limbic and pupillary
    boundaries) be parameterized by center location (xc,yc),
    and radius r (active contour fitting method)
Iris Localization
II. Wildes
 The Wildes et al. system performs its contour fitting in two steps.
  (histogram-based approach)
    First, the image intensity information is converted into a binary


    Second, the edge points vote to instantiate particular contour
      parameter values.
Iris Localization
II. Wildes
 The voting procedure of the Wildes et al. system is
  realized via Hough transforms on parametric
  definitions of the iris boundary contours.
Illustrative Results of Iris Localization

   only that portion of the image below the upper eyelid and above the lower
   eyelid should be included
Rubbersheet Model

                    Each pixel (x,y) is mapped into
                    polar pair (r, ).

                    Circular band is divided into 8
                    subbands of equal thickness for a
                    given angle .

                    Subbands are sampled uniformly
                    in θ and in r.

                    Sampling = averaging over a
                    patch of pixels.
Rubbersheet Model

                    Map Cartesian image coordinates
                     (x, y) to dimensionless polar (r, ө)
                    image coordinates according to
Iris Normalization
 Want to get rid of intensity variation due to iris
 Take mean intensities of 16x16 blocks and subtract
  from image


Iris Normalization
 Finally, perform histogram equalization on 32x32
Feature Extraction

Representation of
iris and also of a

Textured region is
unique for a
Find (nearly circular) iris and create 8
bands or zones
Need to locate the
overall region of
the iris. Then need
to “measure”
texture in 1024
perhaps 128
around each of 8
2-D Gabor Filter
       2-D Gabor filter in polar coordinates:
IrisCode Formation

Intensity is left out of consideration.
Only sign (phase) is of importance.

                                          256 bytes
                                          2,048 bits
Measure of Performance
• Off-line and on-line modes of operation.
Hamming distance: standard measure for comparison of binary strings.

 codeA and codeB are two IrisCodes

     is the notation for exclusive OR (XOR)

 Counts bits that disagree.
Pattern Matching

  The Daugman system computes the normalized
   Hamming distance as

  The result of this computation is then used as the
  goodness of match, with smaller values indicating
  better matches.
Feature Extraction and Matching
 Boles
    Decompose iris into a set of 1D intensity signals
    Apply wavelet transform
    Extract zero-crossing points at different scales
    Compare positions of zero-crossing points for iris
Feature Extraction and Matching
                                    Sample Iris Signature

 Lowest 4 resolution levels   Zero-Crossing representation
  of its wavelet transform    of this signature
CASIA Database
 Used the CASIA Iris Image Database
 108 separate subjects, captured in 2 sessions
    Session 1: 3 images per subject
    Session 2: 4 images per subject

 Total images actively used: 216
CASIA Database
Identification Mode

 Each iris from Session 2 is compared to all irises from
 Session 1. The best match is identified as the correct
Identification Mode
            Representative Example of Identification

                               Step 2: template generated

  Step 1: iris image capture
  and enrolment
                         Step 3: template comparison
                         One to Many (Exhaustive) database match.
                         Never cued
Verification Mode
 Each iris in session 2 is compared one-to-one with
 each iris in session 1
   If distance < threshold, consider it a positive match
   If distance > threshold, reject the match
Verification Mode
            Representative Example of Verification

                                Step 2: template generated

  Step 1: fingerprint capture
  and enrolment
                                   Step 3: template comparison
                                   One to One database match.    4592
                                   Usually cued by PIN
Current Works
Iris Localization
 First, the pupil is
 Compute a rough
  estimate of the pupil
Edge Detection for Pupil

                       Sobel Edge Detector
Change Edge Detection Operator

                     Canny Edge Detector
Morphological Operators

                  a) Original Binary

                  b) Image eroded by a
                     3x3 structuring

                  c) Image dilated by a
                     3x3 structuring
Morphological Operators
Iris Localization
Iris Localization
Iris Localization
Iris Localization
        Future of Iris
1. J. Daugman’s web site. URL:

2. J. Daugman, “High Confidence Visual Recognition of Persons by a Test of Statistical
    Independence,” IEEE Trans. on Pattern Analysis and Machine Intelligence, vol. 15, no.
    11, pp. 1148 – 1161, 1993.

3. J. Daugman, United States Patent No. 5,291,560 (issued on March 1994). Biometric
    Personal Identification System Based on Iris Analysis, Washington DC: U.S.
    Government Printing Office, 1994.

4. J. Daugman, “The Importance of Being Random: Statistical Principles of Iris
    Recognition,” Pattern Recognition, vol. 36, no. 2, pp 279-291.

5. R. P. Wildes, “Iris Recognition: An Emerging Biometric Technology,” Proc. of the IEEE,
    vol. 85, no. 9, 1997, pp. 1348-1363.

6. W. W. Boles, B. Boashash, “A human identification technique using images of the iris
    and wavelet transform”, IEEE Transaction on Signal Processing, vol. 46, no. 4, April

To top