Wavelet Decomposition Feature Parallel Fusion by Quaternion Euclidean Product Distance Matching Score for Finger Texture Verification - Advances in Computational Science and Engineering - page 138

Wavelet Decomposition Feature Parallel Fusion
by Quaternion Euclidean Product Distance
Matching Score for Finger Texture Verification

Di Liu Zheng-ding Qiu and Dong-mei Sun

Institute of Information Science, Beijing Jiaotong University, Beijing, China, 100044
[e-mail]

Abstract. Parallel fusion is a promising fusion method in field of feature
level fusion. Unlike conventional 2-feature parallel fusion with complex
representation, it is tough to represent 4-feature parallel fusion due to
shortage of mathematical significance. To solve this problem, this pa-
per proposes a reasonable interpretation by introducing quaternion. Ini-
tially, this paper defines a novel ROI extraction method for obtaining
more information from middle finger images. After parallel fusion whose
features are extracted by 2D wavelets decomposition coecients from
the same pixel corresponding to 4 separate sub-images, Quaternion Eu-
clidean Product Distance (QEPD), a distance between modulus square
of template quaternion and modulus of tester Quaternion Euclidean
Product (QEP), as matching score, is performed. The scores discrimi-
nate between genuine and impostor of finger texture by threshold effec-
tively. Finally, the experimental result gains a reasonable recognition rate
but at a fast speed. Through a comparison with palmprint QEPD, the
recognition performance of this finger texture QEPD outperforms than
palmprint.

1 Introduction

Nowadays information fusion community is flooded with a large number of bio-
metric fusion approaches. Obviously, information fusion is an information process
utilized to computer technique to automatically analyze information obtained by
sensors serially based on specific criterions, for accomplishing the need of deci-
sion and estimation [1]. At present, we study fusion algorithm on certain specific
context, such as biometric fusion, a novel research topic in information fusion.
Traditionally, in field of the biometric fusion, it is well known that these schemes
are categorized into four major levels, namely (a) sensor level, i.e. data level, (b)
feature level, (c) matching score level, and (d) decision-making level [2,3,4].

(a) Fusion at data level: It is a low-level fusion that data extracted from each
sensor of an arbitrary object are concatenated into a higher dimensional raw data
matrix, e.g. a low unified quality image fusion by multiple channel information,
and then it is used to identification or verification. At the level, many build-up