Wavelet Decomposition Feature Parallel Fusion by Quaternion Euclidean Product Distance Matching Score for Finger Texture Verification - Advances in Computational Science and Engineering - page 146

4.3 Finger Texture Verification

Finger texture verification aims to answer 'whether this person is who he or she
claims to be' [6] by verifying his or her finger texture, which is a one-to-one
matching. If the matching score exceeds the threshold by proposed scheme, it is
accepted as genuine, otherwise, as impostor. The performance of a verification
approach is measured by False Accept Rate (FAR) and Correct Accept Rate
(CAR). They should be made a trade-off by ploting the ROC curve. The ROC
curve is shown on the fig.7.

Fig. 7. ROC curve of the proposed approach

From the figure, the recognition rate goes up with an increase of FAR. The
system keeps a high CAR rate i.e. from 80% to 90%, when FAR is small, e.g.
less than 5%. It achieved 96% of recognition rate since FAR is at 25%.
SVM algorithms have been used as our classifier in this experimental section.
Training set and testing set are totaly separated. In SVM, we assign 10 positive
targets (inner-class) and 10 35 negative targets (outer-class) as training set for

each person (class), so training set for 40 classes is (10 + 350)

×

40 = 14 , 400. No-
tice that the outer-class targets are chosen by each 10 samples from other 35 outer
classes respectively. Finally, the recognition rate approaches 93.36% accuracy.

×

4.4 Speed

The experiment is implemented by Matlab 7.1 on Intel Pentium M processor
(1.6GHz). The execution time of system sections has shown in the table 2. From