18-12-2012, 04:29 PM
Human Identification Using Finger Images
Human Identification Using Finger Images.PDF (Size: 3.32 MB / Downloads: 103)
Abstract—
This paper presents a new approach to improve the
performance of finger-vein identification systems presented in
the literature. The proposed system simultaneously acquires the
finger-vein and low-resolution fingerprint images and combines
these two evidences using a novel score-level combination strategy.
We examine the previously proposed finger-vein identification
approaches and develop a new approach that illustrates it superiority
over prior published efforts. The utility of low-resolution
fingerprint images acquired from a webcam is examined to ascertain
the matching performance from such images. We develop
and investigate two new score-level combinations, i.e., holistic
and nonlinear fusion, and comparatively evaluate them with
more popular score-level fusion approaches to ascertain their
effectiveness in the proposed system. The rigorous experimental
results presented on the database of 6264 images from 156 subjects
illustrate significant improvement in the performance, i.e., both
from the authentication and recognition experiments.
INTRODUCTION
AUTOMATED human identification using physiological
and/or behavioral characteristics, i.e., biometrics, is increasingly
mapped to new civilian applications for commercial
use. The tremendous growth in the demand for more userfriendly
and secured biometrics systems [1] has motivated researchers
to explore new biometrics features and traits. The
anatomy of human fingers is quite complicated and largely responsible
for the individuality of fingerprints and finger veins.
The high individuality of fingerprints has been attributed to the
random imperfections in the friction ridges and valleys, which
are commonly referred to as minutiae or level-2 fingerprint features
[19]. The acquisition of such minutiae features typically
requires imaging resolution higher than 400 dpi. The conventional
level-1 fingerprint features, which illustrate macro finger
details such as ridge flow and pattern type, can be extracted from
the low-resolution fingerprint images. Such features are useful
for fingerprint classification, although the commercially available
automated fingerprint identification systems barely utilize
such level-1 features.
FINGER-VEIN IMAGE PREPROCESSING
The acquired finger images are noisy with rotational and
translational variations resulting from unconstrained (peg-free)
imaging. Therefore, the acquired images are first subjected to
preprocessing steps (see Fig. 3) that include: 1) segmentation of
ROI, 2) translation and orientation alignment, and 3) image enhancement
to extract stable/reliable vascular patterns. Each of
the acquired finger-vein images is first subjected to binarization,
using a fixed threshold value as 230, to coarsely localize the
finger shape in the images. Some portions of background still
appear as connected to the bright finger regions, predominantly
due to uneven illumination. The isolated and loosely connected
regions in the binarized images are eliminated in two steps:
First, the Sobel edge detector is applied to the entire image, and
the resulting edge map is subtracted from the binarized image.
EXPERIMENTS AND RESULTS
In order to ascertain the performance improvement using the
proposed schemes, we performed rigorous experiments on our
own collected database since, to the best of our knowledge, there
is no such database publicly available.
Database
The finger image database employed in this paper consists
of 6264 images acquired from 156 volunteers over a
period of 11 months (April 2009–March 2010) using our
imaging device, as detailed in Section II. In this data set,
about 93% of the subjects are younger than 30 years. The
finger images were acquired in two separate sessions with
a minimum interval of one month, a maximum interval of
over six months, and the average interval of 66.8 days. A
total of 105 subjects turned up for the imaging during the
second (time) session. In each session, each of the subjects
provided six image samples from the index finger to the
middle finger, respectively, and each sample consisting of
one finger-vein image and one finger texture image from
the left hand. Therefore, each subject provided 24 images
in one session.