Image Processing Reference
In-Depth Information
aGrayscaleimage
b Local binary pattern
Fig. 6.6
Local binary pattern example
representation of texture in every neighbourhood [23]. It is important to notice that
representation for every neighbourhood in all blocks is based on the same technique,
but it still produce different results thanks to its dependence on mean values.
6.4.2
Plain CMF
Detection of a plain copy-move forgery is the easiest task for any detection algo-
rithm because the goal is simply to find two equal areas in the image. This refers
to the fact that properties are not changed during the translation of the copied area
to a new location because no transformation (for example, scaling or rotation) is
used. Also, in this case we assume that no post-processing is used (handling with
post-processing is described in Subsect. 6.4.3). Detection task for plain CMF can be
easily solved with the simple 1D CA applied on every block of image [23]:
1. For every pixel
p
c
in the block define neighbourhood
N
for CA as a group of
k
pixels from the row of the image above pixel
p
c
. Pixels are chosen so that one
pixel straight above pixel
p
c
and an equal number of neighbouring pixels from
both sides of
p
c
is selected according to equation (6.6).
N
(
p
c
)=
N
(
p
x
,
y
)=
{
p
x
+
i
,
y
−
1
,
i
=(
−
k
/
2
,...,
k
/
2
)
}
(6.6)
2. For every neighbourhood calculate the mean value using the pixel
p
c
and its
neighbour pixels' intensities. Use a mean value to threshold all pixels' values
p
i
to binary values
b
i
according to equation (6.7).
1
,
p
i
≥
mean
(
N
(
p
c
)
∪
p
c
)
b
i
=
(6.7)
0
,
otherwise
3. Use the fast rule identification method proposed by Sun et al. [22] to gener-
ate a rule in the block that describes the relation between each pixel
p
c
and its
Search WWH ::
Custom Search