Image Processing Reference
In-Depth Information
Table 9.1 Performance indicators for different QP values
Sequence
Image sequence
SSE
R-square
RMSE
Spearman rank
Orbi
Color
0.12
0.97
0.01
0.98
Depth
0.09
0.99
0.01
0.99
Interview
Color
0.06
0.99
0.00
0.99
Depth
0.07
0.99
0.00
0.99
Ballet
Color
0.08
0.95
0.01
0.98
Depth
0.05
0.95
0.00
0.97
Breakdance
Color
0.10
0.97
0.01
0.98
Depth
0.03
0.97
0.00
0.98
All the sequences
Color
1.08
0.94
0.01
0.97
Depth
0.37
0.98
0.01
0.98
for all the sequences in general. Each point of this plot corresponds to the measured
quality of an image frame using both the FR and the proposed method for a given
sequence, compression level, and PLR /( E S / N 0 ). Quadratic approximations using
( 9.14 ) and ( 9.15 ) are also plotted on the scatter plot. Readers should note that these
results are achieved with a few bytes of overhead (to transmit luminance and
contrast statistics of the original image sequence) compared to the FR method.
Individual performance indicators (i.e., SSE, Pearson correlation coefficient
(as R - square ), RMSE, and Spearman correlation coefficient) for the Orbi , Interview ,
Ballet , Breakdance sequences and for all the sequences in general are listed in
Table 9.1 . According to Table 9.1 , it is clear that the proposed Near NR method
(i.e., both “ E - SSIM C ” and “ E - SSIM D ”) is highly correlated with the FR method at all
QP levels, PLRs, and E s / N 0 values considered. This suggests that we can use the
proposed method for quality evaluation of the color image and of corresponding depth
map in place of the FR method with a high accuracy regardless of the sequence type,
compression level, and channel condition being used. This allows developers to use
the proposed method to measure color plus depth based 3D video quality at the
receiver “on the fly” with a very lower overhead for side-information. Based on
the obtained individual quality ratings for color and depth map, receiver-side can
provide feedback to the sender-side about current status of receiving sequences for
system parameter optimization and perhaps receiver can take decisions such as
whether to display 2D content or 3D content. For instance if the measured depth
map quality is significantly low, the quality of the rendered views will be low.
Therefore, the receiver can decide to shift back to 2D video mode instead of 3D video.
In order to better understand the degree of correlation between true 3D percep-
tion (i.e., subjective quality ratings) and results obtained for color and depth map
image quality evaluation with the proposed method, the objective results are
mapped with the subjective MOS results. The mapping between the proposed
Near NR metric for color images (i.e., “ E - SSIM C ” ratings) and subjective quality
ratings for “overall 3D image quality” is shown in Fig. 9.7 . This scatter plot shows
the average subjective quality ( MOS ) and proposed objective quality ratings for the
color image (” E - SSIM C ”) under different PLR /( E S / N 0 ) for all four sequences
Search WWH ::




Custom Search