Iris Recognition Ying Sun AICIP Group Meeting November

Iris Recognition Ying Sun AICIP Group Meeting November 3, 2006 1

Outline • Introduction of Biometrics • Methods for Iris Recognition • Conclusion and Outlook 2

Biometrics Overview • Measures human body features Universal, unique, permanent & quantitatively measurable • Physiological characteristics Fingerprints Face DNA Hand Geometry/Ear Shape Iris/Retina • Behavioral characteristics Signature/gait keystrokes / typing Voiceprint • Example applications Banking, airport access, info security, etc. 3

Advantages of Iris Recognition • Uniqueness Highly rich texture Twins have different iris texture Right eye differs from left eye • Stability Do not change with ages Do not suffer from scratches, abrasions, distortions • Noninvasiveness Contactless technique • High recognition performance 4

Comparison of biometric techniques 5

Verification and Identification • Verification: One to one matching Is this person really who they claim to be? • Identification: One to many matching Who is this person? Identification is more difficult! 6

Identification 10, 000 samples, to identify which one is correct. Suppose being right on an individual test: 0. 9999 To make a correct identification, have to be right on every one of the 10, 000 tests. 0. 9999 10, 000 = 0. 37 Misidentifying: 1. 0 – 0. 37 = 0. 63 63% chance of being wrong! 7

Misidentification increases with the size of database Database of 1, 000 Chance of error: 1, 000 1. 0 - 0. 9999 = 0. 09 Database of 10, 000 Chance of error: 10, 000 1. 0 - 0. 9999 = 0. 63 Database of 100, 000 Chance of error: 100, 000 1. 0 - 0. 9999 = 0. 99995 8

Need Higher Identification Confidence! Iris Recognition Would Satisfy this Criteria. 9

Iris Structure 10

Procedure Employed in Iris Recognition • Iris localization (Segmentation) • Feature extraction • Pattern matching Focusing on Daugman Method 11

Iris Localization Localize the boundary of an iris from the image In particular, localize both the pupillary boundary and the outer (limbus) boundary of the iris. (limbus--the border between the sclera and the iris), both the upper and lower eyelid boundaries Desired characteristics of iris localization: • • • Sensitive to a wide range of edge contrast Robust to irregular borders Capable of dealing with variable occlusions 12

Iris Localization Image Segmentation I(x, y): Raw image : Radial Gaussian *: Convolution The operator searches over the image domain for the maximum in the partial derivative according to increasing radius r, of the normalized contour integral of I(x, y) along a circular arc ds and center coordinates. (active contour fitting method) 13

Feature Extraction • Image Contains Both Amplitude and Phase is unaffected by brightness or contrast changes • Phase Demodulation via 2 D Gabor wavelets • Angle of each phasor quantized to one/four quadrants 14

Gabor Wavelets • • Gabor Wavelets filter out structures at different scales and orientations For each scale and orientation there is a pair of odd and even wavelets A scalar product is carried out between the wavelet and the image (just as in the Discrete Fourier Transform) The result is a complex number 15

Phase Demodulation • • • The complex number is converted to 2 bits The modulus is thrown away because it is sensitive to illumination intensity The phase is converted to 2 bits depending on which quadrant it is in 16

The iris code is a pattern of 1 s and 0 s (bits). These bits are compared against a stored bit pattern. Represent iris texture as a binary vector of 2048 bits 17

Pattern Matching Hamming distance (HD) Calculate the percentage of mismatched bits between a pair of iris codes. (0 -100%) 18

Binomial Distribution • • If two codes come from different irises the different bits will be random The number of different bits will obey a binomial distribution with mean 0. 5 19

Distributions of true matches versus non matches Hamming distances of false matches Hamming distances of true matches If an iris code differs from a stored pattern by 30% or less it is accepted as an identification 20

Probability of the encoding difference between several measurements of the same person P 0 T False acceptance Probability of the encoding difference between different people. False rejection Encoding difference 21 Threshold used to decide acceptance/rejection

Afghan Girl Identified by Iris Patterns 1984 2002 Left eye: HD=0. 24; Right eye: HD=0. 31 22

Summary for Identification • Two codes come from different iris, HD~0. 45 • HD smaller for the same iris • • If the Hamming distance is < 0. 33 the chances of the two codes coming from different irises is 1 in 2. 9 million So far it has been tried out on 2. 3 million test without a single error 23