Digital Signal Processing Reference
In-Depth Information
cxy
denotes color vectors at position
(, )
(, )
xy
:
used in attention calculation,
[,,
T
Lab
;
t
mxy
denotes the average vectors of
tt
(, )
×
neighbor region around
www
(, )
xy
:
[,,
T
Lab
,
t
=
{, ,}
248
w
=
min(
m n
,
)
mn
is width and height
,
,and
cxy
in the Equation(5)
will be replaced by a small neighbor region of position
(, )
(, )
of the image. When the image is polluted by noise,
xy
,such as 5x5
、
7x7.
Thus equation(5) can be changed as follows:
−
−
−
−
Dxy dcxymxy LL aa bb
(, )
=
((, ), (, ))
−
)
2
+
(
−
)
2
+
(
−
)
2
(6)
t
t
m
m
m
2.3
Weighted Color Histogram
Sxy
,)
After color contrast
of each pixel in the image is calculated according to
Sxy
,)
Equation (5),
should be normalized into [0,1] and then be denoted by
Sx
is,
the more important the corresponding pixel is, so all the pixels are weighed by
(, )
Sxy
(, )
Sxy
(, )
(, )
.
can reflect importance of every pixel, the bigger
Sx
and then cumulate to construct weighted color histogram. Equation(2) for
the frequency of pixels in traditional color histogram will be modified as follows:
NN
1
2
Sxyfxyi
(, ) , ,)
i
==
j
NN
11
hi
'
[]
=
(7)
n
1
2
Sxyfxyi
(, ) (, ,)
kii j
== =
11
Obviously, the weighted color histogram can present the significance of colors in an
image well. The more contrast these pixels are with respect to its neighborhood at vari-
ous scales, the greater the visual attention is, corresponding to these pixels the bigger the
weighted value is. In a word , the weighted color histogram can reflect the importance
of pixels position. These are similar to human visual attention mechanism.
3
Spatial Relations in Images
Weighted color histogram includes position information of each pixel and make im-
portant pixels with higher weighted value. But the spatial relations of the pixels with
similar color play also an important role to improve the retrieval performance. Dis-
persed pixels and concentrated pixels like color block may have the same statistic, in
fact they reflect different image information for observer. In this paper, the spatial
relationship of color is considered by the same or similar color distributing cohesion
[12].
