Experimental Results and Discussions
We
have collected some experimental and theoretical test results to measure
success of system.
In
order to get these test results we have used two different databases. In early
development period of the project we have used CASIA [4] database. Two samples
eye image of CASIA database is shown in Fig.17 and two sample eye image of our
own database is shown in Fig.18. The quality of the eye images taken from CASIA
has been very high. We used three training images and one test image. Although
we did not make iris detection we get 72% of success rate. After creating our
own database we had fifteen different person and we used nine training pictures
and one test picture for each person. One of these people was using lens and
two of them have blue colored eye. The recognition success without entering PIN
number is 100% by using first seventeen Eigen faces. We get this success rate
by using Principle Component Analysis (PCA) method. Integration of the hardware
and taking PIN number from user success rate is also 100%.
 
Figure
– 17: Eye samples from CASIA database [4]
 
Figure
– 18: Eye samples from our database
Test
result shows that iris recognition is the most reliable and secure biometric
verification system. Table – 1 shows the verification success for first fifty
eigenface without using keypad. The table clearly shows that using first
fifteen eigenface is enough to verification. Despite first fifteen eigenface is
enough to make verification we are using first twenty eigenface to guarantee
our verification success 100%. Eigenfaces are directly related with the
pictures on the training phase but we are using just first twenty of them.
Although building eigenfaces takes little time (related with size of the
database) verification may guarantee in 3 seconds. Building eigenfaces is
required only if any user removed or registered to system and this process do
not effect the verification run time.
|
# of Used Eigenface
|
Success Rate
|
# of Used Eigenface
|
Success Rate
|
|
1
|
7,11
|
11
|
93,75
|
|
2
|
64
|
12
|
100
|
|
3
|
56,85
|
13
|
100
|
|
4
|
85,33
|
14
|
93,75
|
|
5
|
71,07
|
15
|
100
|
|
6
|
78,17
|
16
|
100
|
|
7
|
78,17
|
17
|
100
|
|
8
|
92,39
|
18
|
100
|
|
9
|
93,75
|
19
|
100
|
|
10
|
93,75
|
20
|
100
|
Table
– 1:
Iris recognition performance test
results.
Iris Verification
Verification test has confirmed our recognition results. We
made our verification test for only 16th user. Test results in Table
– 2 clearly shows Euclidean distance between 16th user and other
users getting regular and uniform while used Eigenfaces are increasing. This
regularity increases verification success rate. If we increase number of Eigen
faces it did not make any difference to make verification more reliable that is
why we are using first twenty Eigen faces.
Table
– 2: Iris verification performance test results.
Note: The values on the table
are multiplied with 10-4
|
eyejudge.com