Image Processing Reference
In-Depth Information
If the tumour/haemorrhage/mammogram cyst or any other region is not
properly visible or if the tumour region has a low contrast, pseudo-colouring
is done to have some knowledge about the tumour region; otherwise, the
greyscale image is kept unchanged. For pseudo-colouring, the RGB (red,
green, blue) image is converted into the CIELab colour image as it is a human
perceptual model. It is a complete colour space describing all the colours that
are visible to the human eye. To convert RGB to CIELAB colour space, the
RGB colour space is first converted to the
XYZ
colour model and then to the
CIELab colour model:
(/)
13
⎛
⎜
Y
Y
⎞
⎟
⎛
⎜
Y
Y
⎞
⎟
>
*
L
=
116
−
16
for
0 008856
.
N
N
⎛
⎜
Y
Y
⎞
⎠
⎛
⎜
Y
Y
N
⎞
⎟
≤
=
903 3
.
for
0 008856
.
⎟
N
⎡
⎢
⎤
⎥
⎛
⎜
X
X
⎞
⎟
−
⎛
⎜
Y
Y
⎞
⎟
a
*
=
500
f
f
N
N
⎡
⎢
⎤
⎥
⎛
⎜
Y
Y
⎞
⎟
−
⎛
⎜
Z
Z
⎞
⎟
b
*
=
200
f
f
N
N
where
16
116
(/)
13
ft t
()
=
,
t
>
0 008856
.
=
7 787
.
t
+
,
t
<
0 008856
.
where
X
n
,
Y
n
and
Z
n
are the CIE tristimulus values with reference to the white
point (
D
65
).
The image is then segmented into several clusters (user defined). It is fol-
lowed by histogram thresholding to eliminate the clusters/pixels which are
not of interest. So, the image contains only the concerned regions. The thres-
holded image is edge detected, which is the boundary of the tumour.
The clustering process uses intuitionistic fuzzy clustering [9]. For grey
image clustering (i.e. without pseudo-colouring), three features are used:
pixel value, mean and standard deviation. For pseudo-coloured image clus-
tering, the CIELab colour model is used where six features are utilized -
three for
L
,
a
* and
b
* values and three for mean values of each
L
,
a
* and
b
*
pixel. Mean is calculated for 3 × 3 neighbourhood of a pixel. A clustered
image contains user-defined segments/regions.
Search WWH ::
Custom Search