![Main specification of this experiment of image quality evaluation [3]](https://www.researchgate.net/profile/Norifumi-Kawabata/publication/303785529/figure/tbl2/AS:839895912304646@1577258198776/Main-specification-of-this-experiment-of-image-quality-evaluation-3_Q320.jpg)
Content uploaded by Norifumi Kawabata
Author content
All content in this area was uploaded by Norifumi Kawabata on Jan 08, 2023
Content may be subject to copyright.
The Fifth International Workshop on Image Media Quality and its Applications, IMQA2011
October 4-5, 2011, Kyoto, Japan
Image Quality Evaluation of 3D CG Images with 8 Viewpoints Lenticular Lens Method
Norifumi Kawabata, Keiji Shibata, Yasuhiro Inazumi and †Yuukou Horita
Graduate School of Science and Engineering, University of Toyama
Gofuku 3190, Toyama-city, Toyama, 930-8555 Japan
E-mail: †horita@eng.u-toyama.ac.jp
ABSTRACT
These days, we have a good chance to watch 3D movies
at home or movie theater. However, there is various
form of stereogram or display style, and these haven’t
ever unified officially yet. In addition, there is no
standardization for evaluating objective video quality of
3D movies. For constructing a better viewing
environment, a suitable viewing condition about the
presence, depth, and nature of the stereoscopic vision
should be known. In this paper, in order to find out the
viewing conditions better, we experimented image
quality evaluation of 3D CG images with 8 viewpoints
lenticular lens method, and considered in detail.
Index Terms ― 3D Images, Computer Graphics, 8
viewpoints lenticular lens method, No Compression,
Image Quality Assessment, Subjective Evaluation,
Absolute Category Rating Method
1. INTRODUCTION
These days, we are in the news that are wide-band
optical communication, super reality system as if we
stayed at home or a movie theater [1], and digital
signage that send much information by using electrical
apparatus, such as display [2].
Now, it’s not decided for 3D method or display
method on the screen because of many kinds of
difference. When we are watching video, it isn’t also
decided definite indicator at official how often
permitting for depth or crosstalk. Until now, they are
studied for many experiments and models of image
quality evaluation, using 3D nature video. However,
there are few research models that require for evaluation
category in 3D CG video [3].
In this paper, we experimented evaluation of 3D
CG images with 8 viewpoints lenticular lens method,
and considered relation of each evaluation category to
hold to what extent presence, depth, nature, and
crosstalk that viewers can permit.
Figure 1 Using image (Wonder World)
Table 1 Main specification of this experiment of
image quality evaluation [3]
Display
Alioscopy 42V model
Image Size
1920x1080(FULL HD)
(pixel)
Image extension
Windows Bitmap
Length of Contents
15 seconds per one pattern
Visual distance
3H
illumination
None
View method
Naked eyes
3D method
Lenticular lens
Evaluation method
ACR method (that is
Single Stimulus
Method[4])
Evaluation people
15 people of University
Students and Graduate
Students
2. PREPARATION OF EXPERIMENT
2.1. Using images and specification
We used images of figure 1 in this experiment that
NICT (National Institute of Information and
Communications Technology) is distributing without
charge [5]. We worked rendering by using Autodesk
Maya 2011 created by Autodesk Corporation, and
changed parameter of “3D to 2D”. This parameter was
changing 8 viewpoints of camera what is called
alioscopy camera (on 3D CG in display used in this
experiment). Yet, if value of parameter is , ranges are 7
types of 0, 0.15, 0.25, 0.35, 0.50, 0.75, and 1.00. Also,
specification of this evaluation experiment is shown
table 1. 8 viewpoints lenticular lens method is used to
display 3D image. This method is prepared 8 viewpoints
sliding constantly (for example, if equals 0.00, space
between two cameras is , and if equals 1.00, space
between two cameras is ) of two images, what is
called left or right eye image included binocular
parallax on the back of lenticular lens, and arranged
mutually by vertical 1 line on a display, and is used lens
which forms half cylinder.
2.2. Evaluation Method
Presenting order of image was repeating of image (15s)
and vote (15s). For evaluation method, we used single
stimulus method prescribed by ITU-T
RECOMMENDATION P.910 to experiment.
Furthermore, evaluation standard were evaluated 5 steps,
those are “very good”, “good”, “fair”, “bad”, and “very
bad”, and calculated MOS (Mean Opinion Score). Also,
evaluation items were 5 types of “Is it presence?”, “Is it
depth?”, “Is it nature?”, “Isn’t it seeing crosstalk?”, and
“How is it general evaluation?” In case of “Presense”,
we evaluates as if we stay in display. In case of “Depth”,
we evaluates if we can see 3D about images. In case of
“Nature”, we evaluates if we see natural condition about
substances in an image. In case of “Not crosstalk”, we
evaluate if we don’t see double images. We added up
results that in advance, we made evaluation members
write questionnaire of “Which evaluation items do you
attach importance watching 3D video by rank?”
Weighting parameter is following equation (1). Still,
weighting of questionnaire’s rank are that first rank is
10 points, forth rank is 1 point. Second rank and third
rank are that difference of weighting parameter become
equal each rank. So, second rank is 7 points, and third
rank is 4 points. From this, we calculated sum of 15
members, and decided weighting of average in sum of 4
evaluation items, that is 1.00. From weighting
percentage of average in sum of 4 evaluation items, I
weighted other ranks.
By the way, is the number of people of rank in each
evaluation items, number of under bar is rank. Also,
is average in sum of 4 evaluation items. shows
about each evaluation items. 4 evaluation items, those
are “Presence”, “Depth”, “Nature”, and “Not Crosstalk”
correspond and . The following equation (2)
calculated general evaluation considered weighting
gained from equation (1).
Table 2 Questionnaire result voted before
experiment by rank
Evaluation
items
1st
2nd
3rd
4th
Sum
Weight
coefficient
(10)
(7)
(4)
(1)
Presence
5
6
2
2
102
1.24
Nature
6
3
3
3
96
1.16
Depth
3
3
5
4
75
0.91
Crosstalk
1
3
5
6
57
0.69
Average
82.5
1.00
(normal)
Figure 2 graph
Figure 3 Weighting coefficient graph
By the way, were parameters of weighting,
was of evaluation items, and were each
evaluation people.
3. EXPERIMENTAL RESULT
3.1 Results of questionnaire of importance for
evaluation category
In advance, results of questionnaire conducted before
experiment, were shown from top, “Presence”, “Nature”,
“Depth”, and “Crosstalk”. Detailed results showed table
2. “nature” is selected the best of all items by evaluation
people, but as a result, “Presence” that second rank is
the best of all, indeed, there are 11 people, was the best
score. On the contrary to this, “Depth” and “Crosstalk”
occupied less than third rank in the majority of all.
Particularly, from table 2, if more than second rank is
important, it found that “crosstalk” isn’t importance for
people of 70% of all.
3.2 Results of evaluation experiment
In this experiment, if was better than 2.50, we
thought that its value could permit. It showed figure 2
that value of gained in experiment by evaluation
items. Error bar, extended top and bottom from plot
points in figure 2 and 3, showed 95% confidential
interval. In “Crosstalk”, as value of parameter came to
increase, it came to decrease. “Presence” shifted within
permitting between 2.5 and 3.0. In “Depth”, got
only lower when value of parameter was 0.00, and after
this, it got without change. In “Nature”, as value of
parameter came to higher, came to lower. In
“General evaluation”, because was less than 2.5
when value of parameter was 0.75 and 1.00, it cannot
permit.
Also, it showed figure 3 that result added up
weighting coefficient showed table 2 in figure 2 of
experimental result. From figure 3, if value of parameter
was more than 0.75, it found that value of general
evaluation converged nearby 2.5. And, from value of
general evaluation, these can classify three types of
good (parameter: 0.00, 0.15, : nearby 4.0), fair
(parameter: 0.25, 0.35, 0.50, : between 3.0 and 3.5),
and bad (parameter: 0.75, 1.00, : nearby 2.5).
4. CONSIDERATION
Seeing From experimental result of figure 2, of
“Crosstalk” decreased more than value of parameter was
0.50. However, it found that other evaluation items
converged in fixed line. We thought that judgment of
other evaluation items was affected because of
increasing crosstalk as parameter came to increase. Also,
although value of parameter approached 1.00, of
“Presence” and “Depth” didn’t increase. We thought
that “Presence” and “Depth” come to offend because of
decreasing of “Crosstalk” and “Nature”. And,
when value of parameter of figure 2 was 0.25 and 0.35,
it’s different to among evaluation items, but
seeing value of general evaluation of figure 3, when
value of parameter are 0.25 and 0.35, it found that
was without change. We thought that when was
0.25 in value of parameter, difference of max and
minimum is small, when was 0.35 in value of
parameter, difference of max and minimum is large,
however, “General evaluation” in was without
change.
5. CONCLUSION
In this paper, we experimented of image quality
evaluation of 3D CG images with 8 viewpoints
lenticular lens method, and considered. From
experimental result, when parameter of each interval of
8 cameras is changed, we could grasp the best was
between 0.15 and 0.25 in value of parameter. Also, we
could catch a feature of graph better by calculating
generate evaluation, and could judge certainly. In this
case, there were two types of method, generate
evaluation voting in evaluation experiment and
experiment result, using weighting coefficient decided
by questionnaire method before experiment. Either can
predict drawing the same lines from graph.
However, in this case, we couldn’t verify about
“crosstalk” and double image. Also, we couldn’t verify
about different images to videos in detail. Included these
points, from now on, we experiment for evaluation of
videos, besides, we advance research.
6. REFERENCES
[1] H. Harajima ITE “Ultra Presece System,” 2010.
[2] K. Muramoto, ITE bulletin Vol.65, No.2, pp.119-120,
“Tendency of Digital Signage,” 2011.
[3] N. Kawabata, K.Shibata, Y.Inazumi, Y.Horita,
Technical report of IEICE, “Video Quality Evaluation of 3D
CG Movies with Active Shutter Glasses,” 2011.
[4]
http://www.ntt.co.jp/qos/qoe/technology/visual/01_5_1.
html, August, 2011.
[5] 3D CG Contents of National Institute of Information
and Communications Technology,
http://3d-contents.nict.go.jp/