Transcript Eyelash and Shadow(ES) Detection

Towards accurate and fast
iris segmentation for iris
biometrics
Zhaofeng He, Tieniu Tan, Fellow, IEEE, Zhenan Sun,
Member, IEEE, and Xianchao Qiu
學生:李仁志
Outline


Introdution
Iris Segmentation Algorithm
• Pupillary and Limbic Boundary
Localization
• Eyelid Localization
• Eyelash and Shadow Detection


Experimental Results
Conclusions
Iris recognition system
How to segment ?

Image information
• Intensity.
• Position.

[0~255]
x,y
Prior knowledge
• Low intensity
• Circular boundary

Daugman
• integrodifferential operator

Wildes
• Hough transform
Iris Segmentation Algorithm
Basic idea of pulling & pushing
method
The pulling & pushing algroithm
’
Center point: O P
Equilibrium length:
(2)
Spring
(3)
Restoring force
(4)
New center point estimation
O
k 1
(x
k 1
,y
k 1
)O F
k
k
(5)
Convergence criterion

Cost function
C (i )  O  O
(i )


( i 1)
(i )
 R R
i 1
(i )
 F  R R
( i 1)
The procedure converges if C(i) is less
than a threshold Cmax
The number of iteration is above a fixed
threshold I max
(6)
The pulling & pushing method with
an illustration
Eyelid localization
Eyelash and Shadow(ES)
Detection
Chi square test
Segmentation results by different algorithms
on one challenging iris image
Experimental results

Iris image database:
• 1. CASIA-IrisV3-Lamp
• 2. ICE v1.0

8-bit intensity images with a
resolution of 640*480
Evaluation Protocol



FAR (False Accept Rate)
FRR (False Reject Rate)
EER (Equal Rrror Rate)
(7)
(8)
Performance Comparison on ICE-Left
iris image database
Performance comparion on CASIA-IrisV3lamp iris image database
ROC curves on ICE-Left iris image
database.
ROC curves on CASIA-IrisV3Lamp iris image database.
Conclusions


We have presented an accurate and
fast iris segmentation algorithm for
iris biometrics
The novel pulling and pushing
procedure is developed to accurately
localize the circular iris boundaries.