Quality Assessment - 3DTV: Processing and Transmission of 3D Video Signals - page 183

respect to the content characteristics and is described as follows:

3 D _ Quality = f 1 ( content ) · image _ quality + f 2 ( content ) · depth _ quality (6.28)

f 1 ( content ) + f 2 ( content ) = 1 (6.29)

6.4.3.1 Weighting Functions

Z-direction (depth direction) motion activity (ZMA) of 3D video is used to
model the weighting functions f 1 and f 2 . For this purpose, a depth map of the
3D video is segmented into blocks of N frames, where N refers to the frame
rate of the depth map, and the standard deviation of each pixel position is
evaluated for each block. After aggregating the standard deviation for all
pixel positions per block the mean standard deviation over all the segmented
blocks is evaluated as shown in Equations (6.30) and (6.31). This is illustrated
in Figure 6.18.

1
N

Y i , j )

1
2

N

t
Y i , j

( Y i , j − μ

t

σ = (6.30)

k = 1

⎡ ⎤

M H W

1
M

⎣

t
Y i , j

⎦

ZMA =

1 σ

(6.31)

t = 1 j = 1 i =

t
Y t , j

t
Y t , j

Y t , j refers to the pixel, σ refers to the standard deviation and μ refers to

the mean of the pixels at i th , j th

position of the block t . N is the number of

frames per block ( N fps), k is the frame index and M refers to the number

of segmented blocks of the depth map. H and W denote the frame height
and width respectively. ZMA is then normalized ( nZMA ) with respect to
resolution of the depth map and the dynamic range of the pixel values, i.e.
2 n

=

1 (255 for 8-bit greyscale representation), where n refers to the number
of bits used for the greyscale representation of the depth map.

−

ZMA

nZMA = (6.32)

(2 n

( W · H ) × − 1)

To model weighting functions f 1 and f 2 , the relative importance of the
subjective ratings (MOS) for image and depth quality with respect to the sub-
jective ratings (MOS) for overall 3D quality are analyzed, for the Interview,
Orbi and Breakdancers sequences, for the quality variations considered in
test 3. Here, the SSCQS method is used instead of the DSCQS method for
subjective testing. This is due to the fact that, in the DSCQS method, the dif-
ference in the MOS between the undistorted original video and the distorted
video, of the same sequence, is considered. Thus, the difference calculation
will eliminate the effect of content characteristics of the video. Finally, f 1 and