1660 N.K. RATHA
et al.
ridge~nes
,Y)
X
Fig. 5. Components of a minutiae feature.
)-
X
window is used to obtain a projection of the pattern in
16 directions. The projection with the maximum vari-
ance is the desired ridge direction for the window. The
result of the enhancement is compared with feature
extraction techniques used in a system currently used
by the U.K. Home office. Performance evaluation is
carried out by comparing features obtained with the
enhancements proposed by this method with the fea-
tures obtained using the available software in the
Home office system. Mehtre 16) computes the direc-
tional image, representing the local ridge direction, in
a block of size 16 x 16 pixels. For this purpose, local
gray-level intensity variances along eight different di-
rections are computed. The direction with the least
variance is the desired ridge direction. A set of eight
7 x 7 convolution masks is applied to the input image
for ridge enhancement. The fingerprint area is seg-
mented from the background before applying stan-
dard locally adaptive thresholding and thinning
operators. Features are obtained based on the compu-
tation of the connection number (CN) described in
reference (11). A postprocessing stage based on a set of
heuristics eliminates the spurious minutiae.
Coetzee and Botha ~rl obtain the ridges by using the
Marr-Hildreth edge operator. This edge map along
with the gray scale image is used to binarize the
fingerprint image. The thresholded image is smoothed
before applying the thinning operation. The direc-
tional image is computed in a fashion similar to the one
described in reference (6). No feature extraction stage is
described. Xiao and Raafat ~9~ assume that the skeleton
image has already been derived from the fingerprint
images. They describe methods to identify spurious
minutiae and eliminate them using the structural defi-
nition of minutiae. For each minutia, statistics of ridge
width and ridge attributes such as ridge length, ridge
direction and minutiae direction are used to decide the
spurious minutiae. Hung ~8~ enhances fingerprint
images by equalizing the ridge widths. The input image
is assumed to be a binary image. Directional enhance-
ment of ridges is done after estimating the local direc-
tion in a small window using a method similar to that
in reference (6). The enhancement process has two
steps: (i) direction-oriented ridge shrinking, followed
by (ii) direction-oriented ridge expanding. The skel-
eton of the enhanced image is obtained by Baja's
algorithm. This paper also describes methods for de-
tecting bridges and breaks as separate features.
The main theme of O'Gorman and Nickerson's tl°~
work is to design filters for fingerprint image enhance-
ment. The k x k mask coefficients are generated based
on the local ridge orientation. Only three orientation
directions are used. Four model parameters derived
from ridge width (Wmax, Wmi,), valley width (l~ma x,
lg'ml,) and minimum radius of curvature are used
to describe a fingerprint. It is assumed that
Wmax q'- Wrain = ]'~Zmax +
['Vmin" The mask is convolved
with the input image. The enhanced image is binarized
and postprocessed. An application-specific integrated
circuit (ASIC) has been designed to meet the comput-
ing requirements of this algorithm. No description of
feature extraction or postprocessing is given.
To summarize, most of the published approaches for
feature extraction use local ridge directions and a
locally adaptive thresholding method. To improve
fingerprint image quality, directional, ridge enhance-
ment is also commonly employed. The thinning step
involves a standard operator. Few published papers
describe a methodology to evaluate the performance
of image enhancement and feature extraction stages.
Often, only portions of the overall feature extrac-
tion module are implemented. Various approaches