Image Processing Reference
In-Depth Information
2
ʼ
x
ʼ
y
þ
C
1
lx
ðÞ
¼
;
y
ð
:
Þ
9
2
ʼ
x
þ ʼ
y
þ
C
1
2
˃
x
˃
y
þ
C
2
cx
ðÞ
¼
;
y
ð
9
:
3
Þ
˃
x
þ ˃
y
þ
C
2
ðÞ
¼
˃
xy
þ
C
3
˃
x
˃
y
þ
Sx
;
y
ð
9
:
4
Þ
C
3
where
x
and
y
are block vectors (i.e., local 8
8 square window) of images
X
and
Y
respectively,
ʼ
x
and
ʼ
y
are the mean of vector
x
and
y
respectively,
˃
x
and
˃
y
are
the standard deviation of vector
x
and
y
respectively,
˃
xy
is the covariance of vector
x
and
y
, and
C
1
,
C
2
, and
C
3
are small constants to avoid the denominators being
zero. Mean and standard deviation are calculated in block based (each block is
8
8). The
luminance
comparison mainly provides a rating for the average
Fig. 9.3 Graphical illustration of the proposed quality metric using the
Ballet
sequence (30th
frame) coded using
QP
30 and transmitted over a network with
PLR
20 %. (a) original depth map;
(b) extracted edge information from the original depth map; (c) original color image; (d) extracted
edge information from the original color image; (e) processed depth map; (f) extracted edge
information from the processed depth map; (g) processed color image, and (h) extracted edge
information from the processed color image