<s>
Video	B-Device
quality	I-Device
is	O
a	O
characteristic	O
of	O
a	O
video	O
passed	O
through	O
a	O
video	O
transmission	O
or	O
processing	O
system	O
that	O
describes	O
perceived	O
video	O
degradation	O
(	O
typically	O
,	O
compared	O
to	O
the	O
original	O
video	O
)	O
.	O
</s>
<s>
Video	O
processing	O
systems	O
may	O
introduce	O
some	O
amount	O
of	O
distortion	O
or	O
artifacts	B-Algorithm
in	O
the	O
video	B-Application
signal	I-Application
that	O
negatively	O
impacts	O
the	O
user	O
's	O
perception	O
of	O
a	O
system	O
.	O
</s>
<s>
For	O
many	O
stakeholders	O
in	O
video	O
production	O
and	O
distribution	O
,	O
assurance	O
of	O
video	B-Device
quality	I-Device
is	O
an	O
important	O
task	O
.	O
</s>
<s>
Video	B-Device
quality	I-Device
evaluation	O
is	O
performed	O
to	O
describe	O
the	O
quality	O
of	O
a	O
set	O
of	O
video	O
sequences	O
under	O
study	O
.	O
</s>
<s>
Video	B-Device
quality	I-Device
can	O
be	O
evaluated	O
objectively	O
(	O
by	O
mathematical	O
models	O
)	O
or	O
subjectively	O
(	O
by	O
asking	O
users	O
for	O
their	O
rating	O
)	O
.	O
</s>
<s>
Such	O
systems	O
encode	B-Application
video	I-Application
streams	I-Application
and	O
transmit	O
them	O
over	O
various	O
kinds	O
of	O
networks	O
or	O
channels	O
.	O
</s>
<s>
The	O
performance	O
of	O
a	O
digital	O
video	O
processing	O
and	O
transmission	O
system	O
can	O
vary	O
significantly	O
and	O
depends	O
on	O
many	O
factors	O
including	O
the	O
characteristics	O
of	O
the	O
input	O
video	B-Application
signal	I-Application
(	O
e.g.	O
</s>
<s>
Objective	O
video	B-Device
quality	I-Device
models	O
are	O
mathematical	O
models	O
that	O
approximate	O
results	O
from	O
subjective	B-Device
quality	I-Device
assessment	I-Device
,	O
in	O
which	O
human	O
observers	O
are	O
asked	O
to	O
rate	O
the	O
quality	O
of	O
a	O
video	O
.	O
</s>
<s>
the	O
packet	B-Protocol
loss	I-Protocol
rate	I-Protocol
on	O
a	O
network	O
and	O
the	O
video	O
coding	O
parameters	O
)	O
are	O
fit	O
against	O
results	O
obtained	O
in	O
a	O
subjective	B-Device
quality	O
evaluation	O
test	O
using	O
regression	O
techniques	O
.	O
</s>
<s>
The	O
terms	O
model	O
and	O
metric	O
are	O
often	O
used	O
interchangeably	O
in	O
the	O
field	O
to	O
mean	O
a	O
descriptive	B-General_Concept
statistic	I-General_Concept
which	O
provides	O
an	O
indicator	O
of	O
quality	O
.	O
</s>
<s>
They	O
can	O
be	O
automatically	O
evaluated	O
by	O
a	O
computer	B-Application
program	I-Application
.	O
</s>
<s>
Full	O
Reference	O
Methods	O
(	O
FR	O
)	O
:	O
FR	O
models	O
compute	O
the	O
quality	O
difference	O
by	O
comparing	O
the	O
original	O
video	B-Application
signal	I-Application
against	O
the	O
received	O
video	B-Application
signal	I-Application
.	O
</s>
<s>
Reduced	O
Reference	O
Methods	O
(	O
RR	O
)	O
:	O
RR	O
models	O
extract	O
some	O
features	O
of	O
both	O
videos	B-Application
and	O
compare	O
them	O
to	O
give	O
a	O
quality	O
score	O
.	O
</s>
<s>
Some	O
of	O
these	O
evaluate	O
specific	O
degradation	O
types	O
only	O
,	O
such	O
as	O
blurring	O
or	O
other	O
coding	B-Algorithm
artifacts	I-Algorithm
.	O
</s>
<s>
However	O
,	O
the	O
picture	B-Algorithm
quality	I-Algorithm
predictions	O
they	O
deliver	O
are	O
not	O
very	O
accurate	O
.	O
</s>
<s>
Hybrid	O
Methods	O
(	O
Hybrid	O
NR-P-B	O
)	O
:	O
Hybrid	O
models	O
combine	O
parameters	O
extracted	O
from	O
the	O
bitstream	O
with	O
a	O
decoded	O
video	B-Application
signal	I-Application
.	O
</s>
<s>
Some	O
models	O
that	O
are	O
used	O
for	O
video	B-Device
quality	I-Device
assessment	O
(	O
such	O
as	O
PSNR	O
or	O
SSIM	B-Algorithm
)	O
are	O
simply	O
image	B-Algorithm
quality	I-Algorithm
models	I-Algorithm
,	O
whose	O
output	O
is	O
calculated	O
for	O
every	O
frame	O
of	O
a	O
video	O
sequence	O
.	O
</s>
<s>
While	O
this	O
method	O
is	O
easy	O
to	O
implement	O
,	O
it	O
does	O
not	O
factor	O
in	O
certain	O
kinds	O
of	O
degradations	O
that	O
develop	O
over	O
time	O
,	O
such	O
as	O
the	O
moving	O
artifacts	B-Algorithm
caused	O
by	O
packet	B-Protocol
loss	I-Protocol
and	O
its	O
concealment	B-Protocol
.	O
</s>
<s>
A	O
video	B-Device
quality	I-Device
model	O
that	O
considers	O
the	O
temporal	O
aspects	O
of	O
quality	O
degradations	O
,	O
like	O
or	O
the	O
MOVIE	B-Device
Index	I-Device
,	O
may	O
be	O
able	O
to	O
produce	O
more	O
accurate	O
predictions	O
of	O
human-perceived	O
quality	O
.	O
</s>
<s>
Metric	O
Usage	O
Description	O
Full-Reference	O
PSNR	O
(	O
Peak	O
Signal-to-Noise	O
Ratio	O
)	O
Image	O
It	O
is	O
calculated	O
between	O
every	O
frame	O
of	O
the	O
original	O
and	O
the	O
degraded	O
video	B-Application
signal	I-Application
.	O
</s>
<s>
PSNR	O
is	O
the	O
most	O
widely	O
used	O
objective	O
image	B-Algorithm
quality	I-Algorithm
metric	O
.	O
</s>
<s>
However	O
,	O
PSNR	O
values	O
do	O
not	O
correlate	O
well	O
with	O
perceived	O
picture	B-Algorithm
quality	I-Algorithm
due	O
to	O
the	O
complex	O
,	O
highly	O
non-linear	O
behaviour	O
of	O
the	O
human	O
visual	O
system	O
.	O
</s>
<s>
SSIM	B-Algorithm
(	O
Structural	B-Algorithm
SIMilarity	I-Algorithm
)	O
Image	O
SSIM	B-Algorithm
is	O
a	O
perception-based	O
model	O
that	O
considers	O
image	O
degradation	O
as	O
perceived	O
change	O
in	O
structural	O
information	O
,	O
while	O
also	O
incorporating	O
important	O
perceptual	O
phenomena	O
,	O
including	O
both	O
luminance	O
masking	O
and	O
contrast	O
masking	O
terms	O
.	O
</s>
<s>
MOVIE	B-Device
Index	I-Device
MOtion-based	O
Video	O
Integrity	O
Evaluation	O
Video	O
The	O
MOVIE	B-Device
index	I-Device
is	O
a	O
neuroscience-based	O
model	O
for	O
predicting	O
the	O
perceptual	O
quality	O
of	O
a	O
(	O
possibly	O
compressed	O
or	O
otherwise	O
distorted	O
)	O
motion	O
picture	O
or	O
video	O
against	O
a	O
pristine	O
reference	O
video	O
.	O
</s>
<s>
VMAF	B-Algorithm
Video	B-Algorithm
Multimethod	I-Algorithm
Assessment	I-Algorithm
Fusion	I-Algorithm
Video	O
VMAF	B-Algorithm
uses	O
four	O
features	O
to	O
predict	O
video	B-Device
quality	I-Device
VIF	O
,	O
DLM	O
,	O
MCPD	O
,	O
AN-SNR	O
.	O
</s>
<s>
The	O
above	O
features	O
are	O
fused	O
using	O
a	O
SVM-based	O
regression	O
to	O
provide	O
a	O
single	O
output	O
score	O
.	O
</s>
<s>
Reduced-Reference	O
SRR	O
(	O
SSIM	B-Algorithm
Reduced-Reference	O
)	O
Video	O
SRR	O
value	O
is	O
calculated	O
as	O
the	O
ratio	O
of	O
received	O
(	O
target	O
)	O
video	B-Application
signal	I-Application
SSIM	B-Algorithm
with	O
reference	O
video	O
pattern	O
SSIM	B-Algorithm
values	O
.	O
</s>
<s>
It	O
is	O
used	O
to	O
evaluate	O
RR	O
entropic	O
differences	O
leading	O
to	O
temporal	O
RRED.It	O
in	O
conjunction	O
with	O
spatial	O
RRED	O
indices	O
evaluated	O
by	O
applying	O
the	O
RRED	O
index	O
on	O
every	O
frame	O
of	O
the	O
video	O
,	O
yield	O
the	O
spatio-temporal	O
RRED	O
No-Reference	O
NIQE	O
Naturalness	O
Image	B-Algorithm
Quality	I-Algorithm
Evaluator	O
Image	O
This	O
IQA	O
model	O
is	O
founded	O
on	O
perceptually	O
relevant	O
spatial	O
domain	O
n	O
natural	O
scene	O
statistic	O
(	O
NSS	O
)	O
features	O
extracted	O
from	O
local	O
image	O
patches	O
that	O
effectively	O
capture	O
the	O
essential	O
low-order	O
statistics	O
of	O
natural	O
images	O
.	O
</s>
<s>
Pedicts	O
score	O
based	O
on	O
those	O
features	O
using	O
SVM	B-Algorithm
.	O
</s>
<s>
An	O
overview	O
of	O
recent	O
no-reference	O
image	B-Algorithm
quality	I-Algorithm
models	I-Algorithm
has	O
been	O
given	O
in	O
a	O
journal	O
paper	O
by	O
Shahid	O
et	O
al	O
.	O
</s>
<s>
The	O
Video	B-Device
Quality	I-Device
Experts	O
Group	O
has	O
a	O
dedicated	O
working	O
group	O
on	O
developing	O
no-reference	O
metrics	O
(	O
called	O
)	O
.	O
</s>
<s>
For	O
a	O
more	O
efficient	O
estimation	O
of	O
video	B-Device
quality	I-Device
in	O
such	O
cases	O
,	O
parametric/bitstream	O
-based	O
metrics	O
have	O
also	O
been	O
standardized	O
:	O
</s>
<s>
Since	O
objective	O
video	B-Device
quality	I-Device
models	O
are	O
expected	O
to	O
predict	O
results	O
given	O
by	O
human	O
observers	O
,	O
they	O
are	O
developed	O
with	O
the	O
aid	O
of	O
subjective	B-Device
test	I-Device
results	I-Device
.	O
</s>
<s>
During	O
the	O
development	O
of	O
an	O
objective	O
model	O
,	O
its	O
parameters	O
should	O
be	O
trained	O
so	O
as	O
to	O
achieve	O
the	O
best	O
correlation	O
between	O
the	O
objectively	O
predicted	O
values	O
and	O
the	O
subjective	B-Device
scores	O
,	O
often	O
available	O
as	O
mean	O
opinion	O
scores	O
(	O
MOS	O
)	O
.	O
</s>
<s>
The	O
most	O
widely	O
used	O
subjective	B-Device
test	O
materials	O
are	O
in	O
the	O
public	O
domain	O
and	O
include	O
still	O
pictures	O
,	O
motion	O
pictures	O
,	O
streaming	O
video	O
,	O
high	O
definition	O
,	O
3-D	O
(	O
stereoscopic	O
)	O
,	O
and	O
special-purposes	O
picture	O
quality-related	O
datasets	O
.	O
</s>
<s>
Some	O
of	O
them	O
have	O
become	O
de	O
facto	O
standards	O
,	O
including	O
several	O
public-domain	O
subjective	B-Device
picture	B-Algorithm
quality	I-Algorithm
databases	O
created	O
and	O
maintained	O
by	O
the	O
as	O
well	O
the	O
.	O
</s>
<s>
However	O
,	O
such	O
a	O
model	O
will	O
be	O
over-trained	B-Error_Name
and	O
will	O
therefore	O
not	O
perform	O
well	O
on	O
new	O
datasets	O
.	O
</s>
<s>
To	O
measure	O
the	O
performance	O
of	O
a	O
model	O
,	O
some	O
frequently	O
used	O
metrics	O
are	O
the	O
linear	O
correlation	O
coefficient	O
,	O
Spearman	B-General_Concept
's	I-General_Concept
rank	I-General_Concept
correlation	I-General_Concept
coefficient	I-General_Concept
,	O
and	O
the	O
root	B-General_Concept
mean	I-General_Concept
square	I-General_Concept
error	I-General_Concept
(	O
RMSE	B-General_Concept
)	O
.	O
</s>
<s>
Other	O
metrics	O
are	O
the	O
kappa	B-General_Concept
coefficient	I-General_Concept
and	O
the	O
outliers	O
ratio	O
.	O
</s>
<s>
Objective	O
video	B-Device
quality	I-Device
models	O
can	O
be	O
used	O
in	O
various	O
application	O
areas	O
.	O
</s>
<s>
In	O
video	B-Application
codec	I-Application
development	O
,	O
the	O
performance	O
of	O
a	O
codec	O
is	O
often	O
evaluated	O
in	O
terms	O
of	O
PSNR	O
or	O
SSIM	B-Algorithm
.	O
</s>
<s>
For	O
example	O
,	O
an	O
IPTV	O
provider	O
may	O
choose	O
to	O
monitor	O
their	O
service	O
quality	O
by	O
means	O
of	O
objective	O
models	O
,	O
rather	O
than	O
asking	O
users	O
for	O
their	O
opinion	O
,	O
or	O
waiting	O
for	O
customer	O
complaints	O
about	O
bad	O
video	B-Device
quality	I-Device
.	O
</s>
<s>
Few	O
of	O
these	O
standards	O
have	O
found	O
commercial	O
applications	O
,	O
including	O
PEVQ	B-Application
and	O
VQuad-HD	B-Application
.	O
</s>
<s>
SSIM	B-Algorithm
is	O
also	O
part	O
of	O
a	O
commercially	O
available	O
video	B-Device
quality	I-Device
toolset	O
(	O
SSIMWAVE	O
)	O
.	O
</s>
<s>
VMAF	B-Algorithm
is	O
used	O
by	O
Netflix	O
to	O
tune	O
their	O
encoding	O
and	O
streaming	O
algorithms	O
,	O
and	O
to	O
quality-control	O
all	O
streamed	O
content	O
.	O
</s>
<s>
It	O
is	O
also	O
being	O
used	O
by	O
other	O
technology	O
companies	O
like	O
Bitmovin	O
and	O
has	O
been	O
integrated	O
into	O
software	O
such	O
as	O
FFmpeg	B-Library
.	O
</s>
<s>
For	O
example	O
,	O
a	O
model	O
that	O
was	O
developed	O
using	O
a	O
particular	O
video	B-Application
codec	I-Application
is	O
not	O
guaranteed	O
to	O
be	O
accurate	O
for	O
another	O
video	B-Application
codec	I-Application
.	O
</s>
<s>
When	O
estimating	O
quality	O
of	O
a	O
video	B-Application
codec	I-Application
,	O
all	O
the	O
mentioned	O
objective	O
methods	O
may	O
require	O
repeating	O
post-encoding	O
tests	O
in	O
order	O
to	O
determine	O
the	O
encoding	O
parameters	O
that	O
satisfy	O
a	O
required	O
level	O
of	O
visual	O
quality	O
,	O
making	O
them	O
time	O
consuming	O
,	O
complex	O
and	O
impractical	O
for	O
implementation	O
in	O
real	O
commercial	O
applications	O
.	O
</s>
<s>
All	O
the	O
visual	O
artifacts	B-Algorithm
are	O
still	O
valuable	O
for	O
video	B-Device
quality	I-Device
.	O
</s>
<s>
It	O
’s	O
often	O
caused	O
by	O
the	O
conversion	O
of	O
24	O
fps	O
movies	O
to	O
a	O
30	O
or	O
60	O
fps	O
video	B-Application
format	I-Application
.	O
</s>
<s>
The	O
main	O
goal	O
of	O
many-objective	O
video	B-Device
quality	I-Device
metrics	O
is	O
to	O
automatically	O
estimate	O
the	O
average	O
user	O
's	O
(	O
viewer	O
's	O
)	O
opinion	O
on	O
the	O
quality	O
of	O
a	O
video	O
processed	O
by	O
a	O
system	O
.	O
</s>
<s>
Procedures	O
for	O
subjective	B-Device
video	I-Device
quality	I-Device
measurements	O
are	O
described	O
in	O
ITU-R	O
recommendation	O
and	O
ITU-T	O
recommendation	O
.	O
</s>
<s>
FFmpeg	B-Library
-	O
FFmpeg	B-Library
is	O
the	O
leading	O
multimedia	O
framework	O
,	O
able	O
to	O
decode	O
,	O
encode	O
,	O
transcode	O
,	O
mux	O
,	O
demux	O
,	O
stream	O
,	O
filter	O
and	O
play	O
pretty	O
much	O
anything	O
that	O
humans	O
and	O
machines	O
have	O
created	O
.	O
</s>
<s>
It	O
is	O
also	O
highly	O
portable	O
:	O
FFmpeg	B-Library
compiles	O
,	O
runs	O
,	O
and	O
passes	O
our	O
testing	O
infrastructure	O
FATE	O
across	O
Linux	O
,	O
Mac	O
OS	O
X	O
,	O
Microsoft	O
Windows	O
,	O
the	O
BSDs	O
,	O
Solaris	O
,	O
etc	O
.	O
</s>
<s>
MSU	O
VQMT	O
-	O
MSU	O
Video	B-Device
Quality	I-Device
Measurement	O
Tool	O
(	O
VQMT	O
)	O
is	O
a	O
program	O
for	O
objective	O
video	B-Device
quality	I-Device
assessment	O
.	O
</s>
<s>
It	O
provides	O
functionality	O
for	O
both	O
full-reference	O
(	O
two	O
videos	B-Application
are	O
examined	O
)	O
and	O
single-reference	O
(	O
one	O
video	O
is	O
analyzed	O
)	O
comparisons	O
.	O
</s>
<s>
EPFL	O
VQMT	O
-	O
This	O
software	O
provides	O
fast	O
implementations	O
of	O
the	O
following	O
objective	O
metrics	O
:	O
PSNR	O
,	O
SSIM	B-Algorithm
,	O
MS-SSIM	O
,	O
VIFp	O
,	O
PSNR-HVS	O
,	O
PSNR-HVS-M	O
.	O
</s>
<s>
OpenVQ	O
-	O
OpenVQ	O
is	O
a	O
video	B-Device
quality	I-Device
assessment	O
toolkit	O
.	O
</s>
<s>
The	O
goal	O
of	O
this	O
project	O
is	O
to	O
provide	O
anyone	O
interested	O
in	O
video	B-Device
quality	I-Device
assessment	O
with	O
a	O
toolkit	O
that	O
a	O
)	O
provides	O
ready-to-use	O
video	B-Device
quality	I-Device
metric	O
implementations	O
,	O
and	O
b	O
)	O
makes	O
it	O
easy	O
to	O
implement	O
other	O
video	B-Device
quality	I-Device
metrics	O
.	O
</s>
<s>
Elecard	O
-	O
Video	B-Device
Quality	I-Device
measurement	O
tool	O
designed	O
to	O
compare	O
the	O
quality	O
of	O
encoded	O
streams	O
based	O
on	O
objective	O
metrics	O
,	O
such	O
as	O
PSNR	O
,	O
APSNR	O
,	O
SSIM	B-Algorithm
,	O
DELTA	O
,	O
MSE	O
,	O
MSAD	O
,	O
VQM	O
,	O
NQI	O
,	O
VMAF	B-Algorithm
and	O
VMAF	B-Algorithm
phone	O
,	O
VIF	O
.	O
</s>
<s>
It	O
provides	O
ways	O
of	O
editing	O
and	O
processing	O
videos	B-Application
.	O
</s>
<s>
VQ	O
Probe	O
-	O
VQ	O
Probe	O
is	O
a	O
professional	O
visual	O
instrument	O
for	O
objective	O
and	O
subjective	B-Device
video	I-Device
quality	I-Device
comparison	O
.	O
</s>
<s>
Vmaf.dev	O
-	O
Vmaf.dev	O
is	O
a	O
tool	O
for	O
video	B-Device
quality	I-Device
analysis	O
that	O
runs	O
on	O
web	O
browsers	O
.	O
</s>
<s>
The	O
tool	O
works	O
with	O
most	O
video	O
container	O
formats	O
and	O
provides	O
per-frame	O
VMAF	B-Algorithm
score	O
visualization	O
.	O
</s>
<s>
QoE	O
prediction	O
in	O
videos	B-Application
is	O
a	O
great	O
challenge	O
because	O
of	O
the	O
multiple	O
situations	O
that	O
may	O
arise	O
and	O
the	O
subjective	B-Device
character	O
of	O
QoE	O
.	O
</s>
<s>
For	O
this	O
reason	O
,	O
to	O
predict	O
the	O
QoE	O
in	O
the	O
most	O
precise	O
way	O
,	O
we	O
have	O
to	O
make	O
use	O
of	O
a	O
good	O
classifier	O
that	O
can	O
detect	O
the	O
most	O
types	O
of	O
errors	O
or	O
unexpected	O
situations	O
that	O
affect	O
video	B-Device
quality	I-Device
.	O
</s>
