Database Reference
In-Depth Information
6.3.2
Edge-Density Measurement
In the previous section, the ECAs are characterized by the degree of similarity (the
CCC
s
value) between the reference and the target images within a predefined region,
R
s
. The resulting ECAs may not be representative of the impression marks, since
images containing no marks can contribute to the high value of
CCC
s
if the two
images are highly correlated. In other words, in the first stage, we only locate
the highly-correlated area on the two images. Here, the second stage requires the
specification of a measurement which describes the notion of a significant mark
area. Edge-density measurement as demonstrated in Chu et al. [
177
] was applied
to accomplish this task. The adoption of the current measurement technique was
motivated by the previous research results showing that a striation mark produced
on the bullet surface image constitutes a significant edge feature. This criterion is
incorporated into the ECA-detection process by using edge density measurement
and applying it to the predefined region,
R
s
.
The edge detection is used to localize the edge in the region
R
s
of the
reference image
. The Canny edge detector [
178
] is used to perform edge
detection. Instead of using a global threshold value specified by the user, the Canny
edge detector adopts the so-called hysteresis thresholding operation, in which a
significant edge is defined as a sequence of pixels with the edge magnitude of at least
one of its members exceeding an upper threshold value, and with the magnitudes
of the other pixels exceeding a lower threshold value. The pixel points judged to
be edges are evaluated as “1”, while other pixel points judged to be non-edge are
evaluated as “0”. The edge density (ED) [
177
] is utilized to describe the ratio of the
number of edge pixel points to the total pixel points of the
n
{
f
s
[
i
,
k
]
}
×
n
reference image
{
f
s
[
i
,
k
]
}
,
number o f pixels at the edges
n
2
ED
s
=
(6.44)
The
ED
s
value quantifies that the region
R
s
contains all detected features on refer-
ence image
. This includes useful features associated with the impressions.
The
ED
s
value is used together with the
CCC
s
value to classify the current region
R
s
as either a relevant or non-relevant area according to a threshold value of
{
f
s
[
i
,
k
]
}
ʾ
EDG
.
Thus, the Eq. (
6.39
) can be rewritten as:
I
=
{
s
:
(
CCC
s
≥
ʾ
COV
)
∧
(
ED
s
≥
ʾ
EDG
)
}
(6.45)
where “
” is the logical AND operation. The remaining process of the extraction
of the ECA then follows Eqs. (
6.40
)-(
6.43
). With this definition, any region
∧
R
s
is regarded as an ECA if the associated sub-images of the reference and the target
images are highly correlated and the sub-image of the reference contains a sufficient
edge feature relative to a predefined threshold value.
Search WWH ::
Custom Search