Wavelet Decomposition Feature Parallel Fusion by Quaternion Euclidean Product Distance Matching Score for Finger Texture Verification - Advances in Computational Science and Engineering - page 144

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

Fig. 4. Finger texture ROI for matching

Table 1. QEPD matching scores among the Fig.4

Fig 4(b) 4(c) 4(d) 4(e) 4(f) 4(g) 4(h) 4(i)

4(a) 0.2449 0.1492 0.3824 0.2015 0.2653 0.4358 0.2419 0.3973
4(b) 0.2021 0.3447 0.2786 0.3139 0.3898 0.2603 0.4234

4(c) 0.2921 0.2674 0.3427 0.4215 0.3001 0.3836

4(d) 0.4082 0.5089 0.4306 0.4570 0.3864

4(e) 0.1777 0.4189 0.1597 0.4165

4(f) 0.3618 0.0722 0.4955

4(g) 0.3314 0.4751

4(h) 0.4771

0.25 to 0.64. For example, fig.4(f) and fig.4(h) belong to the same class because
of a low score of 0.0722. Conversely, fig.4(d) is a impostor to fig.4(f) with a high
score of 0.5089. We will discuss the distribution of genuine and impostor in the
subsection 4.2.

4 Experiment

4.1 Finger Database

BJTU-FI biometric database, an inherited collection work by Institute of In-
formation Science, Beijing Jiaotong University, is utilized for our finger texture
verification experiment. It contains totally 1,500 samples from 98 person's fin-
gers with different illumination conditions. In order to avoid a high dimensional
computation, we acquire middle finger ROI like in the fig.3(b) by our window.
In the test section, we use a subset of the database with 10 samples from each
person, totally 360 images, as our matching set.

4.2 Finger Matching

Sample matching in the database is proposed for investigating the performance of
our scheme. That is, each sample is matched with others samples in the subset