A STUDY ON PERCEPTION OF MOBILE VIDEO WITH SURROUNDING CONTEXTUAL INFLUENCES

donkeyswarmMobile - Wireless

Nov 24, 2013 (3 years and 10 months ago)

203 views

A STUDY ON PERCEPTIO
N OF MOBILE VIDEO WI
TH SURROUNDING CONTE
XTUAL
INFLUENCE
S


Jingteng Xue and Chang Wen Chen

Department of Computer Science and Engineering, State University of New York at Buffalo

Email:
{
jingteng, chencw
}
@buffalo.edu


ABSTRACT


Today people watch mobile videos anytime, anywhere, e.g.
indoors, outdoors, in trains, on streets, etc. The various viewing
contexts significantly impact the viewing experience of mobile
users, but the conventional video quality assessment (VQA)
methods typically disregard the factor of viewing environments.
The main reason lies in the fact that the performance of objective
VQA methods are mostly measured against the subjective VQA
test results while the ITU
-
T recommendations for subjective VQA
me
thods define a set of evaluation conditions that these tests are
required to strictly follow. These conditions, including the viewing
distance, room illumination, display brightness and chromaticity of
background, etc. [1][2], are historically tuned to sim
ulate the living
room television viewing scenario. As a result, the performance of
the conventional VQA schemes still remains unclear in the
application of mobile video quality evaluation. In this paper, we
first distinguish the major factors of viewing co
ntext. A series of
subjective tests are designed and performed to evaluate the
influence of these contextual factors on mobile video. The results
show that (1) perceptual quality is highly correlated with the
context factors; (2) with the presence of the c
ontextual visual
interference, testers spot less video signal distortion and have
lower expectations on video quality. The authors then propose a
VQA model based on Just Noticeable Distortion (JND) theory that
is able to provide context aware prediction of

perceived mobile
video quality and discuss its application in video transmission
systems.


Index Terms


Perceptual visual quality, human visual
system (HVS), just noticeable distortion (JND), mobile video


1.
INTRODUCTION


People enjoy the liberty of watching videos anytime and anywhere.
The massive amount and rapid increase of mobile video user
evidence the
growing

consumption
of
mobile
videos
. It is therefore
important to understand video perception with
in

a mobile context
,
which features time
-
varying wireless channels, limited device
resources and a variety of viewing environments. While the first
two questions are generally considered in the design of mobile
video systems, the prominent aspect of the variety of viewing
co
nditions is not well studied and commonly neglected.

The conventional video quality assessment (VQA) methods
typically disregard the factor of viewing environments mainly
because of the fact that the performance of objective VQA
methods are mostly measured

against the subjective VQA test
results while the ITU
-
T recommendations for subjective VQA
methods define a set of evaluation conditions that these tests are
required to strictly follow. These conditions, including the viewing
distance, room illumination,

display brightness and chromaticity of
background, etc. [1][2], are historically tuned to simulate the living
room television viewing scenario. As a result, the performance of
the conventional VQA schemes remains unclear in the application
of mobile video

quality evaluation
.


1
.1. Related research


There are recent works that propose conducting subjective tests for
mobile videos with real world video streaming services [3][4].
However they do not consider viewing environments as a related
parameter.

Jumisko and Strohmeier have conducted a thorough study on
the impact of the human
-
computer interaction context of mobile
multimedia applications [5][6]. Although their research does not
focus on the visual signal level perception quality, their conclusion
also confirms that the user satisfaction highly correlates with the
context of use.

The influence of the surrounding environment on human
visual perception has been studied with psychophysical
experiments. The influence of display sizes and viewing distanc
es
of mobile devices is introduced and discussed in a recent paper
from Sasse [7]. The influence of the surrounding luminance of
mobile devices is investigated with proposals of color correction
and luminance adjustment [8][9], particularly in the field of

automotive display research. Kim proposed an ambient luminance
adjusted contrast sensitivity function (CSF) in his recent paper
[10]. The influence of temporal frequency on CSF has long been
studied [11][12][13] but the factor in body movements of the
mob
ile user has not yet been studied. In [14], the authors proposed
a JND model to dynamically adapt to different viewing displays.

Although these studies suggest the importance of mobile
video viewing context, to the best of our knowledge there has been
no p
roposal on incorporating the viewing context into mobile video
QoE measurement. It is still an open problem to quantify and
utilize the mobile surrounding context in the application of
perceptual experience evaluation or prediction.


1
.
2
. Contributions of
this paper


The contribution of this paper is in two folds. First, we prove the
influence of mobile context on mobile video perception by
subjective tests and demonstrating the limitation of current VQA
methods. In this paper, we first distinguish the thre
e major factors
of viewing context. We design and perform a series of subjective
tests to evaluate the influence of these contextual factors on mobile
video. The results show that (1) perceptual quality of mobile video
is highly correlated with the mobile
surrounding context factors; (2)
with the presence of the contextual visual interference, testers spot
less video signal distortion and have lower expectations on video
quality. Consequently the context distortion masks the signal
distortion and modulates
the user experience. Moreover, we show
that the conventional VQA schemes present poor performance with
most test contexts because they are not capable to synchronize with
the change of environment.

The second contribution of this paper is that we propose a

new VQA model that is able to provide context aware prediction of
perceived mobile video quality. Based on the Just Noticeable
Distortion (JND) theory, the proposed model employs new
techniques, including spatial frequency translation, luminance
aware CSF

and user motion vector, to simulate the change of
human eye sensitivity against the surrounding context. Therefore
the signal distortion can be masked by the contextual distortion.
The result perceived signal distortion indicates user mobile video
experie
nce. Finally we discuss an application of the proposed
model in video transmission systems.

Our work will benefit the design of mobile video systems in
two ways. First, the context
-
varied subjective test ensures proper
and accurate evaluation of the VQA pe
rformance of video systems.
Second, adapting to the specific surrounding context at the receiver
side maximizes the utilization of the user visual redundancy space,
thereby improving the bandwidth efficiency.


2.

MAJOR CONTEXTUAL FAC
TORS


In this section w
e distinguish the major contextual factors that
influence the perception of mobile video, based on which we will
design a set of context aware subjective tests to evaluate their
influence in Section 3. According to the psychophysical studies
discussed in S
ection 1.1, in the application of mobile video the
viewing conditions can be broken down to the following three
major factors.


2
.
1
. Display size and display viewing distance


When lighting and movements of display are fixed, viewing
quality is mainly dete
rmined by specifications of mobile display,
including picture resolution, display screen size and viewing
distance [7]. For mobile video viewing, the minimum viewing
distance is bounded by the two focus mechanisms of HVS:
convergence and accommodation. Whe
n the viewing distances are
closer than the minimal focal length of human eyes, people will
experience discomfort. On the other hand, the maximal viewing
distance is bounded by the eye accuracy, which is the minimum
viewing angle the eye can resolve. Stati
stical data show that the
preferred viewing distance of mobile video is usually the length of
the arm


about 30 cm.

Display size and viewing distance can be compositely
represented by one variable
-

the viewing ratio (VR), which is
defined to be the ratio

of the viewing distance to the size of the
display screen. Therefore it indicates the viewing angle. Either
width or height can be used to represent the screen size. In this
paper we will use the height. As shown in Table 1, for television
watching in a l
iving room, a typical VR is around four, which
occurs when people watch a 50 cm tall television from a two
-
meter
distance. In the case of tablet video viewing, the VR increases to
about seven. For smartphone viewing, the VR is about ten.


2.2
.
Ambient
luminance


Ambient luminance refers to the illumination surrounding the
viewing mobile devices. The range of ambient luminance is listed
in Table 1.
Section 1.1 introduced the psychophysical research in
this field.

Table 1

List of varied viewing contexts

Viewing Ratio

Device

Monitor

Tablet

Phone

VR

4

7

10

Ambient Luminance

Scenario

Dark

Overcast

Sunlight

Luminance (lux)

100

1000

10000

User Motion

Scenario

Sit

Car

Walk

Acceleration (




)

0.18

0.34

0.83

䕱u楶a汥l琠
|

|



ㄲ1

㌰3


The human eye can
func瑩on in very da牫 and very b物rht
env楲inmen琠but 楴猠con瑲慳琠牥獯汵瑩on 楳 汩m楴ed. The牥fo牥 when
eye adap瑳t 瑯 a b物rh瑥爠 env楲onmen琬 楴 ha猠 瑯 汯獥 con瑲慳琠
牥獯汵瑩on. F楧u牥 1 (b⤠g楶e猠an examp汥l of v楥w楮g 獣牥en w楴h
d楲散琠獵nl楧h琠on i琮 T
he “washout” effect on the video content is
due to the contrast loss and can be simulated as (d).




(a) (b)



(c) (d)

Fig. 1.

Mobile video viewed with different surrounding light
level. (a) in
-
door (b) direct sunlight “washout”. (c) contrast
pe牣e楶ed a猠楮 ⡡⤬ ⡤⤠con瑲慳琠pe牣e楶ed a猠楮 ⡢⤠


䥮 ou爠瑥t瑳Ⱐwe u獥 瑨e 汩gh琠獥n獯爠of 瑨e 瑥t琠sma牴rhone to
e獴sma瑥t 瑨e amb楥
n琠汩gh瑩ng cond楴ionK qhe pam獵ng da污xy pf䤠
mode氠we u獥d fo爠瑥t瑩ng 楳iequ楰ped w楴h a 䍍PSSP 汩gh琠獥n獯rI
wh楣i p牯v楤e猠 four 牥ad楮g猺s NMI NMMI NMMM and NMMMMK puch
quant楺i瑩on ma瑣te猠 瑨e 汯ga物rhm楣i氠 汵m楮ance pe牣ep瑩on
fea瑵牥 of human eye献


2
.
3
.
User movements


Unlike the scenarios of desktop monitor viewing or television
watching, the mobile video playback is frequently accompanied by
body movements of the mobile viewer. Intuitively the intensity of
such movements may not be fierce. However, considering the
smal
l size of the mobile displays, even small movements may
bring significant visual artifacts.

In order to quantify the user motion in different viewing
scenarios, we design and perform a series of field tests. We build
on the Android platform a video player

that constantly records the
readings of accelerometer and orientation sensor during video
playback. Participants watch a video in three different viewing
scenarios: (1) tester sits and holds the mobile display in hands; (2)
tester sits in a moving car and

holds t
he mobile display in hands;

(3)

tester walks and holds the mobile display in hands. The sensor
sample rate is 30Hz.

We first passed t
he collected accelerometer dataset
through

a
low
-
pass filter in order to eliminate the influence of gravity. Then
we removed
the components of acceleration vectors on the z
-
axis,
which is defined as the axis orthogonal to the display plane, such
that the magnitudes of acceleration vector components on the
display plane can be calculated and further smoothed [15]. The
results are shown in Figure 2.



Fig
.

2. User acceleration of three typical scenarios


We calculate the mean acceleration of these scenarios and list
the results in Table 1. In Section 4 these values are translated into
the motion vectors on the image
plane.

In this section we define three contextual factors. For each
factor, we cluster the dynamic range into three categories. In total
we define 27 different contexts.


3.
CONTEXT AWARE SUBJEC
TIVE TEST


In this section we design a set of context aware
subjective tests to
evaluate the influence of the contextual factors we define in
Section 2. We demonstrate the influence of the contextual
interference on mobile video viewing and evaluate the
performance degradation of two conventional VQA methods.


3
.
1
.

Subjective test system design


In order to achieve context
-
varied experiments, we consider two
possible solutions. The first is to simulate different contexts in the
laboratory. This solution provides full control of the entire viewing
process but demands

a substantial amount of resource to achieve
required lighting and motion. We employ the other solution, which
is to collect user ratings with their real world contexts.

Participants watch mobile videos in different contexts on a
daily basis. Whenever the
y watch a sequence, the contextual
information is sensed and recorded. At the end of the playback, the
user is required to submit a subjective score for the viewing
experience, which will be sent to the test server along with the
contextual information. We

prefer this online system over an easy
offline distributed solution because we find it is more expensive to
manage volunteers and test plans offline. Figure 3 shows the block
diagram of the test system we build.




3
.2
.
Test setup and procedures


The tests follow the Single

Stimulus Continuous Quality
Evaluation

(
SSCQE
)

subjective test protocol [1]. For each test, the
tester only watches and rates one video. The distorted videos and
the reference undistorted one are presented in a random order. We
deploy a HTTP video streaming server that contains 150 video
sequences fro
m LIVE video quality database [16][17]. This
database is composed of 10 contents with 15 distorted videos each.
All sequences have a planar YUV 4:2:0 format and 768

432
pixels. We build a database server which holds the user rating and
contextual informati
on from each round of test. The database
server also makes the test plan for each new user by randomly
selecting an order of the 150 sequences. The testers are encouraged
to complete as many tests as possible.

We implemented
the user end player

app
lication

on

the
Android
platform. When a new user logs into the database server,
her test plan is computed. When the user
initiates

a new test, the
application selects a sequence according to the test plan and
downloads from the media server via HTTP. The content
is played
after the entire sequence is downloaded. In this way, the signal and
context distortion is decoupled from the transmission
distortion
.
The typical download time of a sequence is around five seconds.
During playback the

app
lication

record
s
the
acc
elerometer and
light sensor readings at 30Hz
.
When the playback is completed,

the
teste
r is asked to
rate

the quality of the content

on a continuous
quality scale of 0
-
100 (with 0 being bad and 100 being excellent).


3
.
3
.
Test results


We recruited 27 volunteers to participate in our tests and
we have
collected
data from
2950
completed tests. Seven of the testers are
experts in video codecs. All testers have normal vision. All tests
have a viewing ratio of 10. The test devices are Androi
d
smartphones with the same resolution of 800

480. All sequences
were presented in their native resolution.

We define the test context



as a tuple of two factor
variables: the ambient luminance level


(0 for dark, 1 for overcast,
2 for sunlight) and t
he user motion y (0 for sit, 1 for car, 2 for
walk). For example,



refers to the context in which the tester
sits in a dark environment watching a handheld.



is the one
among all tested contexts closest to the laboratory conditions.

The scores from
different testers for the same test video




and the same context



are averaged to yield the resulting mean
opinion score

(






).


refers to the content,


refers to the
distortion method, and




refers to the original sequence of content

.

Media server
Media server
Database server
Database server
Internet
(
via WiFi
/
3
G
)
User
Smartphone
User
Smartphone
Request content
Video
Score
Content parameters
Contextual info
Request content
Video
Score
Content parameters
Contextual info
Accelerometer
Light sensor

Fig
.

3
.

Block diagram of Internet based viewing context aware
subjective evaluation system

F
igure 4 shows the influence of contextual interference on
mobile video viewing by comparing the MOS in various contexts
with MOS with


. It is observed that the
interference

of the
context actually improves the user experience. The major reason is
that the contextual noise masks the signal noise, as several testers
commented

with sunlight or strong movements, it is difficult to
tell the difference among the contents

. Anothe
r observed
phenomenon is that testers usually have a low expectation of
content quality if the contextual interference is strong.

We define


the perceived distortion of a sequence



with
presence of the
contextual

influences



by





(






)


(






)

We evaluate two conventional VQA methods, JND based
DCT domain metric (JND) [13][18] and SSIM based Vim
-
SSIM
fast scheme (VIMSSIM) [19] against the tested contexts. These
full reference VQA algorithms compute the quality diff
erence
between a sequence and its reference as





(



)








Pearson correlation coefficient (PCC)

and
Spearman rank
order correlation coefficient (SROCC)

[20] are used to measure the
correlation between the predicted quality difference



and

the
perceived distortion



over all the nine test contexts. The
results are shown in Table 2. The results of MJND will be
introduced in Section 4.


Table

2 Performance comparison of VQA with context distortion

Context

(xy)

JND

VIMSSIM

PROPOSED

PCC

SROCC

PCC

SROCC

PCC

SROCC

00

0.6002

0.5917

0.8027

0.7891

0.7132

0.7021

01

0.5831

0.5246

0.7831

0.7501

0.7219

0.7103

02

0.4320

0.4179

0.4750

0.4538

0.7009

0.6943

10

0.5933

0.5720

0.6825

0.6431

0.7119

0.7047

11

0.5723

0.5155

0.5305

0.5293

0.7203

0.7087

12

0.4210

0.4173

0.4130

0.4125

0.7192

0.7083

20

0.4776

0.4754

0.4025

0.4017

0.7012

0.6899

21

0.4513

0.4477

0.3924

0.3887

0.7185

0.7161

22

0.4505

0.4489

0.3942

0.3915

0.7043

0.6910


3
.
4
.
Discussion


The results show that (1) perceptual quality of mobile video is
highly correlated with the mobile surrounding context factors. In
particular, the stronger the contextual distortion is, the better the
users feel about the content. (2) With the presence of t
he
contextual visual interference, testers spot less video signal
distortion and have lower expectations on video quality. (3) The
performance of JND and VIMSSIM degrades fast against the
increase of contextual noise. In the sunlight or walking scenario,
t
heir performances are unacceptable.

As a consequence, a context aware VQA scheme is desired.
The fact that the context distortion masks the signal distortion and
modulates the user experience can be utilized in its design.


4.
CONTEXT AWARE VQA MO
DEL


In
this section, we propose a

new VQA algorithm that predicts the
perceptual quality of mobile video with the given viewing
conditions. To achieve this goal, the proposed model needs to
simulate the HVS sensitivity modulation against the viewing
context. We b
uild this model based on the JND theory. The HVS
response to different VR, ambient
luminance

and user motion is
modeled.


4.1 Spatial frequency translation


The most obvious change of CSF responding to
the
change of VR
is the spatial frequency translation. To analyze this, we first
provide

the definition of spatial frequency.























Here



gives the spatial frequency for a given 2D DCT index






.


is the size of DCT block in one

dimension,


is the pixel
size in circles per degree of visual angle obtained by












where


is the physical size of a pixel on the display and


is
the viewing distance. By
a
simple approximation we have





(



)



(



)







where


is the number of pixels of the content. For given
content, spatial frequencies of sub
-
band indices change with, and
only with VR.






















The change of VR leads to the change of spatial frequency for
the same content, and further corresponds to a change in HVS
response, which can be seen as a compression of CSF. Same sub
-
band information shall represent
a
different spatial frequency when
VR
varies and has different HVS sensitivity.



4
.
2
.
Ambient aware
CSF


Kim proposed a
n experimental based

CSF

that is aware of the

ambient luminance in [1
0
]




























































(





)




Fig
.

4.

MOS results with different test contexts

















































Where


refers to
the
ambient luminance in lux,



refers to
the luminance of the device display itself, also in lux.


is
the
physical spatial frequency in cycles per degree.


4
.
3
.
User moti
o
n


First

we compare the body movements and the motion within the
video content. This can be realized with the following formula.






















Here



is the frame rate of the video.


is

the playback time
of a certain frame. If this frame is number


in the display order
then






.





is the instant acceleration of the playback
device on the display plane at time t.



is a virtual motion
vector which can represent the instant de
vice movement




.


is
the pixel density of the screen (number per meter). The calculated
magnitudes of virtual MVs corresponding to the mean
accelerations for the three viewing scenarios discussed previously
are listed in Table
1
.

Below we introduce the

perceptual influence of temporal
frequency and the scheme to incorporate content motion as well as
viewer body movement into temporal frequency computation.
Temporal frequency of a video signal refers to the rate at which the
image varies. It relates to b
oth
the
motion information and
the
spatial frequency. The relationship can be shown as:














where


is the spatial frequency which is translated to
the
designated VR prior to further process.



is the velocity of motion
in degree/s on the retina (
eye) plane. To obtain

, movements of
the

eye and
the
visual target should
both
be considered.

Human eyes
pursue the moving video

object

at the speed


,
which is bounded by [11]:



















where


is the gain of the smooth pursuit
eye movements
with the empirical value of 0.98.



is the minimum eye velocity
due to the drift movement and is about 0.15



.




is the
maximum velocity of the eyes. It is normally 80




[
11
]. The
n
the

velocity


can be calculated by:















Here



is the velocity on the image (display) plane
, which
consists of the motion of the content and the motion of the device.
Defin
ing



to be the overall motion (in pixel) on the display plane,
we have














Here

is the motion vector of the content (macro
-
block),
which is
extracted from the stream
.

The computation of


relies on real time motion extraction
and incorporation. Considering the common decoding delay,


should be a predicted value. Based on this consid
eration, and also
for the sake of complexity and power constraint, we approximate
the device acceleration


to be non
-
real
-
time but predicted from
the physical readings in past periods. Now the display plane
velocity can be
obtained
:
















4
.
4
.
VQA model


With


,




and CSF

being
calculated, the
conventional JND model
can be applied. The sub
-
band domain just noticeable distortion
threshold for frame

, block

, sub
-
band index






is






















































































































































and








refer to luminance adaptation
(intra pictur
e) and contrast masking effect, which are identical to
[13]. Note that the final threshold












is a function of

,
ambient luminance


and user motion

.

Giving the target sequence



and the reference sequence


,
we compute their sub
-
band domai
n difference


|





|

and
the JND threshold


of


. The perceptual distortion of a block
is the result of a weighted pooling [21] of the index differences.









(







































)





The perceptual distortion of a frame is derived from a pooling
of all blocks within this frame. Given the fact that heavy local
distortion will degrade the
quality

of the entire picture, the pooling
leans to the local maximum distortion [18].








(






)





(






)

Similarly, the perceptual distortion of the video is derived
from a pooling of all frames.





(





)





(





)

Here




are constant parameters.


4
.
5
.
Test results


We computed the performance of the proposed VQA algorithm on
the previous
data set. The motion information and ambient
luminance recorded during the tests were used as the input. The
results are shown in Table 2. Although in the context of low
contextual distortion, the proposed scheme is not as accurate as
VIMSSIM, but it track
s the change of context and maintains a
good performance over all tested contexts while JND and
VIMSSIM become unacceptable quickly along the increase of the
contextual interferences.


5.
DISCUSSION AND CONCL
USION
S


This paper first enumerates the related
factors of
mobile
viewing
context

and discusses their influence on perception experience
based on an analysis of human vision model. Second we propose a
novel paradigm of subjective video quality evaluation that involves
surrounding parameters as test vari
ables. Our preliminary test
results evidence the strong correlation between viewing condition
parameters and the QoE of mobile video. Last we introduce a
viewing context aware VQA scheme. The test results show

the
proposed VQA approach

is able to synchroni
ze with the change of
viewing context.

Our work will benefit the design of mobile video systems in
two ways. First, the viewing condition involved subjective test
ensures proper and accurate evaluation of the QoE performance of
video systems. Second,
the
proposed VQA model can be utilized to
achieve a context aware perceptual optimized video
adapt
ation
system and
improve bandwidth efficiency.

The proposed contextual related perception model is both of
theoretical guidance and practical implications. The ri
ch set of
sensors on handh
e
lds today enables these smart devices to
accurately understand mobile users’ viewing circumstances. The
mobile device can be intelligently equipped to respond to the
change of environment. A viewing context powered adaptation
sys
tem can then optimize
the
viewer’s QoE by tailoring video
more properly.

This research contributes significantly to the QoE design for
mobile video system by introducing and utilizing the information
of end user environment. In the future, we plan to apply

the main
ideas from this research to the design of QoE driven video delivery
system to maximize the benefit of user environment adaptation.


6
. REFERENCES


[1]

“Methodology for the subjective assessment of the quality of
television pictures,” ITU
-
R
BT.500
-
12, 2009.


[2]

“Subjective video quality assessment methods for multimedia
applications,” ITU
-
T Rec. P. 910


[3] K
. Wac,

S.
Ickin
,

J.
Hong,
L.
Janowski,
M.
Fiedler
,

and
A.K.
Dey, “Studying the experience of mobile applications used in
different cont
exts of daily life,”
ACM SIGCOMM
W
orkshop on
Measurements
U
p the
S
tack (W
-
MUST '11).


[4]

I.
Ketyko,
K.
Moor,
T.
Pessemier,
A.
Verdejo,
K.
Vanhecke,
W.

Joseph,
L.

Martens, and
L.
Marez, “QoE

measurement of mobile
YouTube video streaming,” In Proceedings of the 3rd workshop on
Mobile
V
ideo
D
elivery (MoViD '10).


[5]

S.

Jumisko
-
Pyykkö

and

T.
Vainio, T
,

“Framing the context of
use for mobile HCI, review paper about mobile contexts of use
between

2000
-
2007”, Intern
ational Journal of Mobile
-
Human
-
Computer
-
Interaction (IJMHCI), 3(4).


[6]

S.

Jumisko
-
Pyykkö,
and T.

Utriainen, “A hybrid method for
the context of use: evaluation of user
-
centered quality of
experience for mobile (3D) televis
ion”, International Journal of
Multimedia Tools and Applications: Special

issue on Mobile
Media Delivery
.


[7]

M.A.Sasse and H. Knoche, “The big picture on small screens
delivering acceptable video quality in mobile TV,” ACM
Transactions on Multimedia Com
puting, Communications and
Applications, vol. 5, 2009.


[8]

C. Liu and M.D. Fairchild, “Measuring the relationship
between perceived image contrast and surround illumination,”
IS&T/SID 12th Color Imaging Conference, 282
-
288, 2004.


[9]

S. Y. Choi, M. R. Lu
o and M. R. Pointer, “The Influence of the
relative luminance of the surround on the perceived quality of an
image on a large display,” Proc. 15th Color Imaging Conference,
157
-
162, 2007.


[10]

Y. J. Kim
,

An automatic image enhancement method
adaptive to
the surround luminance variation for small

sized
mobile transmissive LCD,”

Consumer El
ectronics, IEEE
Transactions on
, vol.56, no.3, pp.1161
-
1166, Aug. 2010


[11]

D. H. Kelly, “Motion and vision. II. Stabilized spatiotemporal
threshold surface,” J Opt Soc
Am, vol. 69, pp.1340

1349, 1979


[12]

Y. Jia, W. Lin, and A.A. Kassim, “Estimating just noticeable
distortion for video,” IEEE Transactions on Circuits and Systems
for Video Technology, vol. 16, no. 7, pp. 820

829, July 2006.


[13]

Z. Wei and K. N. Ngan, “
A temporal just
-
noticeable distortion
p
rofile for video in DCT domain,”

Im
age Processing,
15th I
EEE
International Conference on
, pp.1336
-
1339, 12
-
15 Oct. 2008


[14]

J. Xue and C. W. Chen
, “
Towards viewing qual
ity optimized
video adaptation,”

Multimedia and

Expo (ICME), 2011 IEEE
International Conference on , vol., no., pp.1
-
6, 11
-
15 July 2011


[15]

Android Developer Manual,
http://developer.android.com


[16]

K. Seshadrinathan, R. Soundararajan, A. C. Bovik and L.

K.
Cormack, “Study of subjective and objective quality assessment of
video”, IEEE Trans on Image Processing, vol.19, no.6, pp.1427
-
1441, June 2010.


[17]

K. Seshadrinathan, R. Soundararajan, A. C. Bovik and L. K.
Cormack, "A subjective study to evaluate v
ideo quality assessment
algorithms", SPIE Proceedings Human Vision an
d Electronic
Imaging, Jan. 2010


[18]

F. Xiao, “DCT
-
based video quality evaluation”


[19]

C. Vu and S. Deshpande, “ViMSSIM: from image to video
quality assessment,” ACM Workshop on Mobile

Video 2012


[20]

Video Quality Experts Group (VQEG), “Final report from the
video quality experts group on the validation of objective models
of video quality assessment,” Phase II 2003.


[21]

J. Xue and C. W. Chen, “A new perceptual quality metric for
vi
deo transrating for mobile devices,” In Proceedings of the 2010
ACM multimedia workshop on Mobile cloud media computing
(MCMC '10).