Information Technology Reference
In-Depth Information
(
)
I
n
= α
(
l
,
m
,
n
)
⋅
CD
l
,
m
,
n
(3)
.
The cumulative depth factor represents the proportional effect of depth perception
for every packet under the assumption that significant focal attention of the viewer
is received by nearer objects in the video scene as compared to distant objects. It
reduces the importance level of distant objects to give priority to nearer objects. It
is calculated as follows.
D
⎧
,
if
D
〈
D
⎪
⎨
av
(
)
⎪
⎩
100
,
,
=
CD
l
m
n
av
th
(4)
1
otherwise
where
D
av
is the average depth of the pixels in that packet. The depth threshold va-
riable,
D
th
prioritizes the content according to its depth information. The optimal
value for this variable is related to the average depth of the whole video frame,
which would differ from sequence to sequence. However, it has been observed
from experiments that pixels with depth values less than 70 do not contribute sig-
nificantly to perceptual quality; hence a threshold value of 70 would be a good
estimate for sub optimal separation of video packets. It must be noted that depth
information represents the distance of a pixel from the viewing point calibrated in
gray scale levels. The scale values range from 255 for nearest point to 0 for far-
thest point. The distortion estimation model is discussed in the next section.
3.4 Optimized Transmission and Decoding
This section describes the estimation of the distortion due to corruption of data in
a video packet during its transmission.
The video quality is affected by the quantisation distortion
E
(
Q
D
) and channel
distortion. Channel distortion is composed of concealment distortion and distor-
tion caused by error propagation over frames coded using temporal prediction.
Concealment distortion depends on the concealment techniques applied at the de-
coder. If errors in the video data are concealed with concealment data from the
previous frames, then it is called temporal concealment and distortion caused by
such a scheme is called as temporal concealment distortion
E
(
D
T
). Errors can
propagate in two ways, either in the temporal direction or in the spatial direction.
Frame to frame error propagation through motion prediction and temporal con-
cealment is called temporal domain error propagation
f
tp
. The distortion estimation
model used in this work has similarities with the method proposed in [30]. How-
ever, the distortion induced due to error propagation is calculated differently and
an adaptive intra refresh technique [31] is used to minimize the error propagation
factor, where a fixed number of macroblocks in every P frame are randomly se-
lected and 'intra' coded. Previous research has suggested that spatial error con-
cealment is less important than the temporal error concealment [32] so in this
work only temporal concealment is considered.
