Information Technology Reference
In-Depth Information
(a)
1
0
−4
10
−4
10
−4
⋅
⋅
⋅
2500
2500
2500
156
156
156
5
10
15
5
10
15
5
10
15
(b)
9.77
9.77
9.77
0.610
0.610
0.610
0.038
0.038
0.038
5
10
15
5
10
15
5
10
15
1
0
−4
10
−4
10
−4
⋅
⋅
⋅
2500
2500
2500
156
156
156
5
10
15
5
10
15
5
10
15
9.77
9.77
9.77
0.610
0.610
0.610
0.038
0.038
0.038
5
10
15
5
10
15
5
10
15
10
−4
1
0
−4
10
−4
⋅
⋅
⋅
2500
2500
2500
156
156
156
5
10
15
5
10
15
5
10
15
9.77
9.77
9.77
0.610
0.610
0.610
0.038
0.038
0.038
5
10
15
5
10
15
5
10
15
1
0
−4
10
−4
10
−4
⋅
⋅
⋅
2500
2500
2500
156
156
156
5
5
10
15
5
10
15
9.77
9.77
9.77
10
15
0.610
0.610
0.610
0.038
0.038
0.038
5
10
15
5
10
15
5
10
15
Fig. 12.7
Co-occurrence matrices
a
calculated for an example person image
b
for each channel
(HSV—from
left
to
right
) and for each direction (0
ⓦ
,45
ⓦ
,90
ⓦ
, 135
ⓦ
—from
top
to
bottom
)
12.4.6 Edge Histogram Descriptor
Edge Histogram Descriptor is a technique which is used to characterize image texture
(
EHD
). It is one of the methods defined within the MPEG-7 standard [
36
]. It describes
the texture of a grey-scale image in the form of a histogram. This descriptor extraction
process is calculated as follows [
41
].
First, similarly to CLD extraction, a preprocessing stage, in which the image is
re-scaled and rotated accordingly, is carried out. Next, the analysed image is divided
into 16 regions (4
×
4). Afterwards, each region is additionally divided into blocks
of 4 pixels (2
2). For each block, a dominant texture directionality is estimated
utilizing a set of filters (Fig.
12.8
).
To estimate the main directionality, the strength is calculated for each of its types
according to the equation:
×
3
m
]
=
a
k
(
i
,
j
)
×
f
]
(
k
)
(12.10)
[
dir
[
dir
k
=
0
where
m
is the strength of a specific direction,
a
k
is the
k
th block element at
[
dir
]
(
i
,
j
)
coordinates and
f
]
(
k
)
corresponds to
k
th filter coefficient.
[
dir
Search WWH ::
Custom Search