Upload
peregrine-baldwin-jefferson
View
216
Download
0
Embed Size (px)
Citation preview
Comparison of subjective test methodologies
VQEG Berlin meeting June 2009
P. Le Callet, R. Pépion
Context, methodologies and issues
Context:
HRCs (coder, processing, transmission …)
Resolutions
Applications and services
ACR (5 , 11 categories …)
Pair ComparisonSAMVIQ
DSCQS
The value (e.g. accuracy, stability) of protocols might depend on the context … and the targeted goals
Outline
• Study 1: ACR 5 vs SAMVIQ HD H264• Study 2: Preference Tests vs SAMVIQ
«processing»• Study 3: ACR5 vs ACR 11 vs SAMVIQ encoded +
processing• Study 4: ACR5 for encoded + transmission error
Study 1: ACR 5 vs SAMVIQ HDMotivations: HDTV high quality in a short range=> quality measure should be precise and discriminative
Absolute Category Rating (ACR)
- random order
- only one viewing
- category scale
- no explicit reference
...Good
Subjective Assessment Methodologyfor Video Quality (SAMVIQ)
- user-driven order
- multiple viewing (natural?)
- continuous scale
- explicit reference
5
Previous and new studies
[Brotherton, 2006] correlation on CIF (352x288):
CC(MOSACR, MOSSAMVIQ) = 0.94
New studies: - Resolutions: QVGA, VGA and HD 1080i50 (viewing distance according to the resolution)- HRC: coding artefacts only (H264 AVC and SVC)
CC(MOSACR, MOSSAMVIQ) =
HDTV
VGA
QVGA 13°
19°
33°
0.969
0.942
0.899
6.73
9.31
14.06
visualfield
RMSDiff=
ACR and seems to provide “equivalent” resultsup to a certain resolution
6
Accuracy vs Number of observers
5 10 15 20 25 3002468
101214
SAMVIQACR'
confi
denc
e in
terv
al
number of observers
24
« Suitable methodology in subjective video quality assessment: a resolution dependent paradigm » Stéphane Péchard, Romuald Pépion and Patrick Le CalletProceedings of the Third International Workshop on Image Media Quality and its Applications, IMQA2008, Chiba, Japan, September 2008
Outline
• Study 1: ACR 5 vs SAMVIQ HD H264• Study 2: Preference Tests vs SAMVIQ
«processing»• Study 3: ACR5 vs ACR 11 vs SAMVIQ encoded +
processing• Study 4: ACR5 for encoded + transmission error
Study 2: Preference Test vs SAMVIQ « processing »Motivations: HDTV pre post processing, comparison between format on a 1080p display= > No other impairments
1080p SRC
Pre Processing(interleaced and down Scaling)
1080i, 720p
Pre Processing(deinterleaced + down Scaling)
720p
Post Processingdeinterleaced + up Scaling)
Post Processing(up Scaling)
1080p PVS
Study 2: some results
0
10
20
30
40
50
60
70
80
90
100
Référenceexplicite1080p50
référencecachée
1080p50
1080i50TDeint
720p(source
1080p50)Lanczos
720p(source
1080p50)NN
720p(source
1080i50)Lanczos
720p(source
1080i50)NN
ancrebasse
MO
S
SAMVIQ
-3
-2,5
-2
-1,5
-1
-0,5
0
0,5
1
1,5
1080i50 TDeint 720p50 (source1080p50)Lanczos
720p50 (source1080p50) NN
720p50 (source1080i50)Lanczos
720p50 (source1080i50) NN
720p50 natif
Pré
fére
nce
Preference Test1080p SRC compared to other PVS7 categories preference test
Generally good agreement but…further analysis is required(Thurstone Mosteller, CI …)
Outline
• Study 1: ACR 5 vs SAMVIQ HD H264• Study 2: Preference Tests vs SAMVIQ
«processing»• Study 3: ACR5 vs ACR 11 vs SAMVIQ encoded +
processing• Study 4: ACR5 for encoded + transmission error
Study 3: ACR5 vs ACR 11vs SAMVIQ encoded + processing
Motivations:Comparison of 1080p50 with other HD and SD formats on a 1080p display =>
compression + processing
Compression:H264 coderAll formats (e.g. 1080p or i, 720p …) are coded at 3,6 and 9Mb/s and
decoded before post processing.
Processing:All formats are displayed in 1080p50 after decoding
1 deinterlacer : Smooth (VirtualDub/Avisynth),2 Upscalers : Bilinear and Lanczos (VirtualDub/Avisynth).
Study 3: PVS generation
1080i50
720p50
1280x1080i50
1280x1080p50
SD
Deint
Upscale 1
Upscale 2
DeintUpscale 1
Upscale 2
DeintUpscale 2
Upscale 1
Upscale 2
29 HRC(8x3 HD+2x2 SD+1Ref)
x
3 SRC
=
87 PVS
3Mb/s
6Mb/s
9Mb/sNot forSD
Upscale 1
ACR5 vs ACR11: correlation
correlation between ACR 5 and 11: 0.98
5 contents
1
1.5
2
2.5
3
3.5
4
4.5
5
0 2 4 6 8 10
MOS ACR 11 levels
MO
S A
CR
5 l
evel
s Canal
équivalence
DepartCross
FootRennes
Manege
Stockholm
Study 3: SAMVIQ vs ACR11, PVS generation
X
1080i50
720p50
1280x1080i50
1280x1080p50
SD
Deint
Upscale 1
Upscale 2
DeintUpscale 1
Upscale 2
Deint Upscale 1
Upscale 1
Upscale 2
10 HRC(8HD+1SD
+1Ref)
x
2 SRC
=
20 PVS
Study 3: ACR11 vs SAMVIQ (on 20 PVS)
• Good correlation between ACR and SAMVIQ (0.97) => may be questionnable for high quality score
4 contents
0
10
20
30
40
50
60
70
80
90
100
0 2 4 6 8 10
MOS ACR 11 levels
MO
S S
AM
VIQ
Canal
équivalence
DepartCross
FootRennes
Manege
Study 3: score distribution
ACR5
ACR11
0
10
20
30
40
50
60
70
0 10 20 30 40 50 60 70 80 90 100
SAMVIQ
Study 3: CI distribution
Outline
• Study 1: ACR 5 vs SAMVIQ HD H264• Study 2: Preference Tests vs SAMVIQ
«processing»• Study 3: ACR5 vs ACR 11 vs SAMVIQ encoded +
processing• Study 4: ACR5 for encoded + transmission error
Study 4: ACR5 encoded + transmission error
• The goal : analyse the relation between the position of the transmission error and the MOS on SD sequences.
• Each content is coded at 4 or 6Mb/s and some simulation of transmission errors are tested.
• Advanced FEC and Error concealment technique (ROI based)
Study 4: ACR5 encoded + transmission error
X
14HRC
(Trans-MissionErrors)
=84
PVS
Study 4: ACR5 encoded + transmission error
Reminder: coding only (study 3)