31.07.2013 Views

IRIS RECOGNITION BASED ON HILBERT–HUANG TRANSFORM 1 ...

IRIS RECOGNITION BASED ON HILBERT–HUANG TRANSFORM 1 ...

IRIS RECOGNITION BASED ON HILBERT–HUANG TRANSFORM 1 ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

636 Z.Yang,Z.Yang&L.Yang<br />

x<br />

c1<br />

c2<br />

c3<br />

c4<br />

c5<br />

c6<br />

r<br />

250<br />

200<br />

150<br />

(a) (b)<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

10<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−10<br />

0<br />

10<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−10<br />

0<br />

200<br />

180<br />

50 100 150 200 250 300 350 400 450 500<br />

160<br />

0 50 100 150 200 250 300 350 400 450 500<br />

x<br />

c1<br />

c2<br />

c3<br />

c4<br />

c5<br />

c6<br />

r<br />

200<br />

150<br />

100<br />

50<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

20<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−20<br />

0<br />

10<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−10<br />

0<br />

10<br />

0<br />

50 100 150 200 250 300 350 400 450 500<br />

−10<br />

0<br />

150<br />

50 100 150 200 250 300 350 400 450 500<br />

100<br />

0 50 100 150 200 250 300 350 400 450 500<br />

(c) (d)<br />

(e) (f)<br />

Fig. 11. (a) Iris image with bright illumination; (b) the same iris image with dark illumination;<br />

(c) the EMD decomposition result of the 10th line of (a); (d) the EMD decomposition result of<br />

the 10th line of (b); (e) the result by removing all the residues of lines from (a); (f) the result by<br />

removing all the residues of lines from (b).<br />

To show the robustness of our method to noise, we added 20 dB Gauss white<br />

noise to the iris image as shown in Fig. 12(a) and calculate the main frequency<br />

centers and the energies of 18 orientations in I1 of the original normalized image<br />

and the noise normalized image, as shown in Figs. 12(b) and 12(c), respectively.<br />

It can be seen that most features of the noisy iris image just have small changes<br />

compared with those of the original iris. Therefore, the proposed feature is robust<br />

to high frequency noise.<br />

3.4. Iris matching<br />

After feature extraction, an iris image is represented as a feature vector of length<br />

108. To improve computational efficiency and classification accuracy, Linear Discriminant<br />

Analysis (LDA) is first used to reduce the dimensionality of the feature<br />

vector and then the Euclidean similarity measure is adopted for classification. LDA<br />

is a linear statistic classification method, which intends to find a linear transform T<br />

as such that, after its application, the scatter of sample vectors is minimized within<br />

each class, and the scatter of those mean vectors around the total mean vector can<br />

be maximized simultaneously. Further details of LDA may be found in Ref. 8.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!