Iris Biometrics Flashcards
(15 cards)
Iris Recognition Strategy
Find outer boundary
Find inner boundary
Obtain region
Obtain Code
Iris Recognition Problems
Lighting
Eyebrows
Peculiarities
Eyelids
Iris Recog Advantages
Uniqueness
Stable
Non Contact
First Iris Recog Approach
Near Infra Red Images are used because they reduce reflections and reveal clear iris texture.
Detecting the Iris
A mathematical operator searches for circular edges in the image, to find the pupil and iris boundaries.
The function searches the image for the maximum blurred (gaussian) partial derivative over all radii r and centre positions x0,y0.
Extracting the Iris Region
Region extraction and polar to Cartesian coordinate transformation - uses rubber sheet model.
This standardises pupil shape despite pupil dilation, eye rotation or head movement.
Encoding the Iris Pattern
Local regions of an iris are projected onto ‘quadrature 2D Gabor Wavelets’.
Each wavelet produces a complex number which define a phasor in the complex plane.
The angle of the phase is quantised to one of four quadrants, each with a unique 2 bit code.
This section of the iris is given this 2 bit code and the process is repeated.
Hamming Distances for Iris Recognition
Hamming distance is used to measure how different two strings are by counting how many bits are different.
Can calculate iris dissimilarity as normalised masked XOR between codes.
Millions of tested irises produced binomial distribution, almost all had a HD of 0.5 (random match). Very unlikely that two different iris are considered similar.
Quantile-Quantile (Q-Q Plot)
Observed cumulative vs Predicted Binomial Hamming Distances. If the results follow a perfect binomial distribution, the plot should be a straight line (x=y)
Iris Recognition Performance
In ideal conditions (same camera, distance, lighting), matching irises will have very low hamming distance peaking at 0, all different irises should hove around 0.5 HD.
In unfavourable conditions, the spread of HDs for the matching iris will be greater (0-~0.33) but the HDs for different Iris should stay the same.
Iris recognition in twins
Genotype (DNA) vs Phenotype (The displayed characteristic)
Twins have the same DNA but different Irises.
Genetically identical eyes also roughly followed binomial spread of HDs
Iris Recognition Deployment
UK Immigration
Access control in Japan
Children’s Identification and
Location Database (CHILD)
UAE watch list
Afghan Refuges
Airports
Military
Criminal
Iris Databases
Chinese Academy of Sciences (CASIA 1 and 3)
Iris Challenge Evaluation ICE
University of Bath
Multimedia University
West Virginia University
Using Texture to detect fake irises
Fake irises have sharper edges and smoother texture compared to the course natural texture of live ones.
Grey-level co-occurrence features measure how often different pixel intensities appear next to each other and can detect fake irises.
Iris Quality Assessment
Coarse Localisation of the pupil
Fourier Transform of two local regions
The mean of two local quality descriptors
SVM based decision
