Wavelet Decomposition Feature Parallel Fusion by Quaternion Euclidean Product Distance Matching Score for Finger Texture Verification - Advances in Computational Science and Engineering - page 148

finger texture QEPD, so as to affect recognition performance. Under the same
classifier condition, we propose a comprehensive comparison for two QEPDs as
follows.

Table 3. A comparison of palmprint QEPD and finger texture QEPD

Palmprint QEPD Finger texture QEPD

Original image size 128 × 128 12 × 72

Wavelet decomposition level 3 2

Quaternion matrix size 16 × 16 3 × 18

Recognition rate 92.87% 93.36%

This table shows a difference between palmprint QEPD and middle finger tex-
ture QEPD. As to original sample size, palmprints in our database are acquired
as low resolution 128 × 128 images, while finger texture images are 12 × 72,
whichislessthanpalmprint's.Thismeans system needs more time to process
original palmprint images for wavelet decomposition. On the item of wavelet
decomposition level, we here make a choice of level 3 and 2 respectively. Harr
kernel is chosen in the decomposition. Here decomposition at Level 2 is faster
than level 3. That is to say, this can save time for finger texture, rather than
palmprint. Notice that quaternion matrix size is referred as to the size of matrix
that stores PP or . Evidently, this size of palmprint is larger than finger

texture's. It means palmprint QEPD can hardly save time cost as finger texture
QEPD. Finally, the recognition rate of finger with 93.36% is higher than that of
palmprint. That is, both time cost and recognition performance of finger texture
palmprint outperform than palmprint. From data, we can conclude that trait of
finger texture is a better biometric than palmprint in QEPD.

|PQ|

5Con lu on

This paper proposes a novel approach 2D Wavelet Decomposition Feature Paral-
lel Fusion with a QEPD matching score for middle finger texture verification. As
4-feature parallel fusion by wavelets decomposition, the paper first defines a novel
finger ROI, finger texture, for wavelet decomposition feature extraction. Then it
gives an interpretation by introducing quaternion, in order to avoid shortage of
mathematic significance since fusion. Then the scheme defines QEPD based on
a term of quaternion QEP, as matching score. Finally, the experimental result
is capable of gaining a better recognition rate 93.36% than that of palmprint
but at a fast speed. Through this comparison, the size of original finger texture
image is smaller than palmprint's. Meanwhile, an intermediate in the algorithm,
quaternion matrix has a lower dimensionality than palmprint. Therefore, it is
safe to conclude that trait of finger texture is a better biometric than palmprint
in QEPD.