Information Technology Reference
In-Depth Information
According to the analysis on the relationship between the perceived stereo-
scopic quality and distortion conditions, the influence of different distortion types
and image contents on the perceived quality is different. Subsequently, we will
simply analyze the performance of PSNR on the quality prediction for stereo-
scopic images. Figure 6 and Figure 7 show the mean of DMOS values and 95%
confidence interval plotted against PSNR (dB) from each distortion type for the
Art
image and
Bowling
image, respectively. As expected, the PSNR values in-
crease as DMOS values decreasing for both images. However, it is found that the
difference of the DMOS values of the
Art
image between the Gaussian blurring
and JPEG2000 compression is less than that of the
Bowling
image for the same
PSNR value. This phenomenon indicates that the performance of PSNR is de-
pendent on the distortion type and image content, and it might not be an appropri-
ate metric for evaluating the quality of stereoscopic images. Therefore, we will
propose more suitable objective quality metrics for the stereoscopic image based
on some 2D image quality metrics and the disparity information. The extensive
analysis on the experimental results above is a fundament work and provides a
solid basis for designing objective quality metrics.
100
JPEG Compression
Gaussian blurring
JPEG2000 Compression
white Gaussian noise
90
80
70
60
50
40
30
20
10
0
15
20
25
30
35
40
45
Peak Signal to Noise Ratio(PSNR)(dB)
Fig. 7
Scatter plot of
Bowling
image: DMOS values versus PSNR
3 Perceptual Stereoscopic Image Quality Metric Based on 2D
Image Quality Metrics and Disparity Analysis
Accurately predicting stereoscopic quality is an important issue for improving the
ability and feasibility of compression and transmission schemes for stereoscopic
images. Although many 2D image quality metrics (IQMs) have been proposed that
work well on 2D images, developing quality metrics for 3D presentations is almost
an unexplored issue. As indicated in Section 2, PSNR is not appropriate for evalu-
ating the quality of stereoscopic images. Therefore, in this section, we will intro-
duce some well-known 2D IQMs and investigate their capabilities in stereoscopic
