Display Content Change Rate Analysis for Power Saving in
Transmissive Panels
Raghu Bankapur
1
, Krishna Kishor Jha
2
and Vaisakh P. S.
3
1
Technical Lead, Multimedia & Systems Division, Samsung R&D Institute India Bangalore Pvt. Ltd, 16/1, 8th Cross,
Muni Reddy Layout, Mahadevapura, 560048, Bangalore, Karnataka, India
2
Technical Lead, Multimedia & Systems Division, Samsung R&D Institute India Bangalore Pvt. Ltd, #303, Thirunaga
Wings, 2nd Cross, Byrasandra, C V Raman Nagar, 560093, Bangalore, Karnataka, India
3
Technical Manager, Multimedia & Systems Division, Samsung R&D Institute India Bangalore Pvt. Ltd, 2870, Phoenix
Building, Bagmance Constellation Business Park, Doddenakundi, 560037, Bangalore, Karnataka, India
Keywords: Backlight Control, Color Distribution Coefficient, Content Change Rate Coefficient, Luminance, Panel
Tuning Lookup Table, Self-Organizing Map.
Abstract: Energy conservation to protract battery life is a major challenge for sustained richer user experience in
mobile devices. This paper presents Content Change Rate Coefficient (CCRC) and image Color Distribution
Coefficient (CDC) based method to govern backlight scaling and image luminosity adjustment to reduce
power consumption in backlight module. The existing methods intending to achieve similar results have
significant shortcomings like inter-frame brightness distortion, flickering effects, clipping artifacts, color
distortion. These problems are addressed in proposed technique and user's visual perception is not
compromised while ensuring image fidelity (less than 6.2% degradation) and reduces nearly 80mA current
consumption. Proposed method derives Aggressiveness Coefficient from color distribution and content
change rate to address above problems. Method is prototyped on Samsung Galaxy Tab AS and Galaxy
Mega2 and experimental results clearly show 16% reduction in current consumption and 88% reduction in
flickering artifacts, which is active phone usage time getting extended by ~1hr for an average usage of
20hrs.
1 INTRODUCTION
Users of portable consumer electronics products,
such as mobile phones, tablets, GPS etc. reckon to
operate these devices for a longer period of time. So
these product manufacturers always strive towards
constantly optimizing their power consumption to
meet this objective.
Transmissive panel is one of the most common
Liquid Crystal Display (LCD) screen, which
requires a backlight as the light source and there is
no reflective film in the back of LCD screen. These
types of LCDs are preferred over other display
technologies in wide range of consumer products
because of low cost, manufacturing ease, light
weight, high resolution, good color performance and
other advantages. However, there are some
downsides i.e. low image contrast ratio and constant
power consumption irrespective of image displayed.
In typical LCDs, the image is displayed by
controlling the LC transparency with constant
backlight luminance. The image contrast ratio for
such display panels is impoverished because the
light is not blocked completely at the lowest gray
level. In LCD based products power is primarily
attributed to backlight, nearly 90% (Carroll and
Heiser, 2010); because the backlight always operates
with the constant luminance regardless of input
image and content based backlight control is
considered as one of the most effective technique for
power saving in LCD panels. Several methods are
proposed in the literature (Tsai et al., 2009;
Kerofsky and Daly, 2006; Chen et al., 2013; Choi et
al., 2002; Raman and Hekstra, 2005; Cheng et al.,
2004; Hsiu et al., 2011) that leverage this and reduce
power consumption. However, they are prone to
multiple artifacts (Hsiu et al., 2011) like inter-frame
brightness distortion, frequent flickering effects,
clipping at saturated pixels, color distortion etc. due
to irresolute operation.
446
Bankapur, R., Jha, K. and S., V.
Display Content Change Rate Analysis for Power Saving in Transmissive Panels.
DOI: 10.5220/0005733404460456
In Proceedings of the 5th International Conference on Pattern Recognition Applications and Methods (ICPRAM 2016), pages 446-456
ISBN: 978-989-758-173-1
Copyright
c
2016 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
The present paper proposed a unique method to
effectively overcome these defects comparable to
conventional methods, while maintaining the richer
user experience. Most importantly, proposed method
conserves power. In particular we make below three
contributions addressing following research
questions.
First, how to minimize inter-frame brightness
distortion and frequent flickering effects? A Content
Change Rate based Coefficient (CCRC) is derived to
restrict frequent backlight modifications to reduce
flickering effects and inter-frame distortion by 88%.
This is presented in section 3.1, 4.2, 5.2.
Second, how to minimize color distortion and
clipping at saturated pixels? An image classification
model to classify image into color rich and simple
UI with 93% accuracy is modelled. The pixel level
modifications to displayed image are based on this
classification to ensure minimum color distortion.
This is presented in section 3.2, 5.1.
Third, how to maximize power saving without
compromising user experience? A content change
rate and color distribution coefficient based
algorithm is modelled to achieve high power saving
when content on the display is Simple UI (not color
rich) and less dynamic. This is presented in section 4.
User study shows that more than 25% of content on
display is less dynamic. Also the present algorithm
saves 16% power with more than 93.8% Structural
SIMilarity (SSIM) (Wang et al., 2004) accuracy.
The experimental results for proposed method are
discussed in section 5.
Figure 1 shows block diagram of proposed
technique. It consists of revised display subsystem
with an additional content change and color
distribution analysis block and backlight and image
intensity control block. Content change analysis and
color distribution analysis block quantifies Content
Change Rate Coefficient (CCRC) and Color
Figure 1: Block diagram for proposed technique.
Distribution Coefficient (CDC) respectively and
other block modulate image intensity and backlight
level based on CCRC and CDC.
2 LCD AND PERCEIVED
LUMINANCE
Before stating algorithm, it is necessary to
understand relation between brightness, backlight,
power consumption and intensity of image. For LCD
based products, major power consumption is
attributed to backlight and it is the main source for
brightness of the LCD, which is measured in lux.
There are different techniques to control brightness
of display such as backlight and intensity.
The optical light flow in LCD subsystem (Raman
and Hekstra, 2005) is illustrated in Figure 2, where
in backlight is main source of light (Backlight
Luminance BL) and fragmentary directed light pass
through pixel depends on intensity value.
Figure 2: Transmissive property of TFT LCD.
The perceived intensity of an image is denoted by
(1)
I = βLY
f
r
(1)
Where β is the transmittance of LCD Panel, L is
backlight luminance and Yfr is the average
luminance value of the frame (Raman and Hekstra,
2005). ‘L’ can be controlled linearly by Backlight
Level (BL) and the ‘Yfr’ can be controlled by
increasing/decreasing the Luma component for the
frame called Image Intensity Level (IL). Concurrent
adjusting of intensity and backlight level in proper
way leads to energy conservation, but just modifying
these above mentioned parameters may cause
defects in displayed image.
So there are two parameters that contribute to the
final luminance and hence, same brightness
perception can be produced by multiple
combinations of backlight and pixel’s intensity value.
Multiple experiments were done to find relation
between perceived brightness, backlight level, image
intensity along with differences in power consumed.
Figure 3 shows brightness variation w.r.t backlight
Display Content Change Rate Analysis for Power Saving in Transmissive Panels
447
level at multiple intensity of image. From this figure,
it’s evident that same luminance level can be
achieved from a range of backlight levels and
modifying image intensity. For instance, brightness
value of 100 lux can be achieved either by backlight
level 250 or 130 with intensity increase from 0(I 0)
to 25(I 25).
Figure 3: Brightness Vs Backlight Level at different
Intensity level.
Power consumption variation with respect to
increase in backlight level and intensity of image is
further explained in figure 4. While power
consumption is directly proportional to backlight
values, increase in pixels intensity has negligible
impact, less than 5mA. For instance, reducing
backlight level from 250 to 130, there is around
110mA power reduction. And as illustrated in Table
1 for luminance values of 100, 75 and 50,
considerable power can be saved by manipulating
pixel intensity and backlight simultaneously,
ensuring similar perceived brightness. For instance,
100lux luminance intensity can be produced from
device by any pixel intensity and backlight level
combinations like (250, 0), (205, 4), (170, 8) etc. And
all these combinations drain different amount of
currents. Like, by reducing backlight level from 250
to 170 and increasing intensity value from 0 to 8,
75mA current consumption is reduced without
degrading image perception to the eye.
It is shown that controlling backlight and
intensity concurrently lead to power benefit,
however these techniques suffer from drawbacks
like inter-frame brightness distortion, flickering
effects, clipping artifacts, color distortion etc. This
present work tries to redress these associated
drawbacks and suggests a novel method of content
change rate and color distribution based backlight
and intensity control. Significance of content change
and color distribution analysis is discussed in next
section.
Figure 4: Power consumed vs Backlight level at different
intensity level.
Table 1: Luminance at different intensity and Backlight
level.
Luminance
Backlight
Level(BL)
Intensity
Level(IL)
Current
consumption
100
250 0 330
205 4 280
170 8 255
75
210 0 305
175 4 258
150 8 235
50
150 0 235
130 4 220
100 8 190
3 KEY CONCEPTS IN
PROPOSED METHOD
This section discusses about two of the main
inventions done during this research work. First one
is display content analysis which is prime discovery
done to avoid defects like inter frame flicker and
distortion and second is avoidance of color distortion
for color rich images. Next sub sections discuss
detail about these inventions.
3.1 Display Content Change Analysis
and Its Significance
Display content adjustment and backlight control is
done dynamically based on image characteristics.
This is because every image has different color
distribution and any such adjustment of pixels
intensity in saturated images will lead to quality
degradation and clipping artifacts.
In conventional techniques, for use cases where
screen is changing at 30fps or continuously updating,
intensity and backlight adjustment coefficient is
calculated on per frame basis and on every frame
ICPRAM 2016 - International Conference on Pattern Recognition Applications and Methods
448
backlight is adjusted by different amount. As a result,
flicker and inter frame distortion become prominent
and thus resulted in bad user experience.
The study on the use cases where Display
content is static with over 50 users was done. It was
found that, on an average in more than 25% of use
time, frame is idle or display updates at low
frequency. Also in these use cases limited color
variations are present. Like during call when speaker
mode is ON, or while reading Pdfs books,
messaging etc., display is static for majority of time
and also contents have limited color distributions. In
these cases per frame backlight control is not
required and hence flickers and distortion can be
avoided by gradual backlight adjustment.
Figure 5 shows display content change analysis
for different application installed on smartphones
obtained from study performed. It was found that for
major applications like launcher, contacts, browser,
email etc. screen was static for more than 20% of
time. So, selectively implementing this algorithm in
use cases limited by static condition can still have
huge power saving together with rich user
experience.
Figure 5: Statistics of Display content change rate. Note
that this data taken when there are no frame submissions
for duration of 2s.
Display content change analysis can be explained
in terms of Content Change Rate (CCR) that is
derived based on frame submission rates to display.
For use cases like video play, gaming applications
CCR will be high as content in such cases changes at
high rate and incase of launcher, contacts, call etc.
CCR will be low. Hence content change rate
analysis is done to calculate CCR intensity based on
which image intensity adjustment and backlight
control is performed.
There are cases wherein displayed images are
having saturated colors, thus change in intensity may
cause image distortion and also in such cases
intensity has negligible effect on luminance. So in
proposed technique, detecting such use cases and
applying final intensity based on color distribution
analysis is shown; discussed in next section.
3.2 Color Distribution Coefficient
(CDC)
Proposed algorithm attempts to minimize the image
distortion in case of images which are color rich, and
at the same time tries to save more power on image
which are simple UI. Thus image on display is
classified into two types- Color Rich and Simple UI.
Color Rich images have wide color distribution
whereas Simple UI consist few colors. For instance
UI of Alarm Clock, Contacts Application, etc. are
Simple UI images and photos/videos are color rich
images.
Self-Organizing Map (SOM) (Schatzmann and
Ghanem, 2003; Kohonen, 1990) based technique is
used for this classification. The SOM is
unsupervised neural network based dimension
reduction technique. Each node in the SOM has a
weight vector whose dimension is equal to the
dimension of input data. Rectangular SOM can be
visualized as a rectangular mesh of n x m nodes (e.g.
Figure 10 shows 3 x 3 rectangular SOM). The length
of line connecting the nodes is Euclidean distance
between the weights of Nodes. SOM exhibits a
property that it maintains the original data topology,
i.e. similar items get mapped to nearby nodes. As a
result of this topology preservation, when SOM is
trained with the input data it tries to take the shape
of data distribution. For example consider an
imaginary image whose pixel values, when plotted
in 3D scatter plot, the plot takes the shape of sphere.
If a rectangular SOM is trained with this image pixel
values as input, the rectangular SOM mesh plot will
take the shape of sphere too. This property of SOM
is harnessed here to make the color distribution
estimation. A 6 x 6 rectangular SOM is created
which is trained with pixel values of input image.
Training algorithm for the SOM is described as
below (Kohonen, 1990).
1. Initialize the Weight vectors W(I, J) with random
values.
2. Determine the winner neuron Wc(t) for input
pixel x of the image using Euclidean distance
formula.
3. Update the weight vectors using (2)
W
i
(t+1) = W
i
(t) + α(t)h
ci
(t)[x(t) – W
i
(t)] (2)
Where α is learning rate and
h
ci
is a neighborhood
Gaussian function given by (3)
Display Content Change Rate Analysis for Power Saving in Transmissive Panels
449
h
ci
= exp(−
|
|

|
|

(
)
)
(3)
In above equation
|
|

|
|
is the distance
between the position(x, y) of winning node and i
th
node in the rectangular SOM map.
The neighborhood radius (o(t)) and learning rate
(α (t)) also decays with time, with below decay
function (4).
L(t) = L
0
exp(-
)
(4)
Where N is total iteration count, and t is current
iteration number.
4. Repeat steps 2 and 3 until total iteration count is
reached.
Fig. 8 shows the 3D scatter plot of color pixel values
(RGB) in Color rich image (Fig. 6). It also shows the
SOM (black mesh) which is obtained after training.
It can be seen how the SOM has taken the shape of
input distribution (green dots). Also a similar map is
constructed for a Simple UI image (Fig. 7) shown in
Fig. 9. It can be seen in SOM of Fig. 9 it is mostly
concentrated in one region and considerably lesser
stretched as it has less number of colors. Whereas in
Figure 8, the SOM is stretched more showing that
the distribution of color in Fig. 6 is more. A similar
observation is made with a set of 1000 images in
which 500 were simple UI ones- will be discussed in
results section.
Figure 6: Color Rich Image. Figure 7: Simple UI.
Figure 8: SOM for Colo
r
rich Image.
Figure 9: SOM for Simple UI.
Based on this, it is inferred that the stretch of the
SOM correctly represents the color distribution in
the image. This observation of difference in Color
distribution and hence the spread of SOM, is used as
an objective criteria to classify Color rich and
Simple UI images. The Color Distribution Metric
(CDM) can be derived from (5)
=




D
L
(I,J) + D
T
(I,J)+ D
B
(I,J)+
D
R
(I,J)
(5)
Where D
L
,D
B
,D
R
,D
T
are the Euclidean distance of
weights in SOM from node (I,J) in SOM map from
left, top, bottom and right node respectively. Figure
10 shows the location D
L
,D
B
,D
R
,D
T
of nodes in
SOM.
Figure 10: 3 x 3 Rectangular SOM showing position of
D
L
,D
B
,D
R
,D
T
nodes of (5).
CDM represents the amount by which the SOM
map is spread and hence a high value of CDM
implies color rich image and whereas low represents
less number of colors and hence Simple UI.
CDC is derived from CDM using (6), T
is the
CDM threshold for Color Rich image. The value of
T
is determined experimentally as it depends on
the resolution of the image.
CDC =
1–

≤
0>
(6)
Performance: In order to reduce the running time of
the proposed CDC calculation the display image is
scaled down to 180x320 size. Since the primary
purpose is to determine the color distribution scaling
down the image still retains its color property.
Average running time for CDC computation was
measured to be 7ms.
This CDC value is used in the algorithm
explained in next section to compute the
aggressiveness of the algorithm.
ICPRAM 2016 - International Conference on Pattern Recognition Applications and Methods
450
4 PROPOSED METHOD
As explained in section 2, same luminance can be
achieved at lower power consumption by
concurrently reducing backlight level and increase in
intensity. This section discusses about proposed
algorithm that explains when and by what amount
the intensity and backlight values need to change. It
also explains use of Color Distribution Coefficient
(CDC) and Content Change Rate (CCR) to minimize
the artifacts occurring due to change in backlight
and intensity level.
Figure 11 shows algorithm flow for proposed
technique. Algorithm starts with loading panel tuned
look up table when device boots. On frame update
Content Change Rate Coefficient (CCRC) and CDC
is computed. User is provided with authority to set
Aggressiveness (

) based on power saving
requirement. During this research we invented
Effective aggressiveness (

) formula which is
function of CCRC, CDC and 

. Based on
effective aggressiveness new IL and BL are
calculated and applied to system. In following sub-
sections each of these blocks are explained in detail.
4.1 Generating Panel Tuned Look-up
Table (LUX_TABLE)
The method to create lux table is depicted in Figure
12. Creating LUX_TABLE involves below two
steps-
Step 1: Display uniform gray image on display for
tuning brightness table. Gray Image is taken for
readings since RGB distribution is uniform in gray
color image.
Step 2: Measure Luminescence (Lux) for all
backlight levels from bl=0 to bl=255 and Intensity
Levels varying from il=0 to il=25 and fill
LUX_TABLE [255][25] table accordingly. For
backlight level bl=0 and Intensity Level il=0
measured lux value LUX_TABLE [0] [0] is filled in
the table as shown in Table 2.
Table 2: Lux mapping table.
Intensity
Backlight
il=0 il=1 il=25
bl=0
LUX_TAB
LE [0][0]
LUX_
TABLE
[0][1]
LUX_
TABLE
[0][25]
…… ….. ……. ….
bl=255
LUX_
TABLE
[255][0]
LUX_
TABLE
[255][1]
LUX_
TABLE
[255][25]
Figure 11: Flowchart for proposed technique.
Figure 12: Generating LUX_TABLE.
Setup for generating Panel Tuned look-up table
consists of Display Color Analyzer (Konica Minolta
CA-310) and a laptop with tuning tool installed. Test
device is kept in closed environment (Black box) to
avoid surrounding luminance interference as shown
in Figure 13.
Tuning tool enables mapping different backlight
levels and intensity levels and Color analyzer
enables measurement of amount of light coming out
from test device. Luminance recorded from color
analyzer is tabulated to use in proposed algorithm.
Display Content Change Rate Analysis for Power Saving in Transmissive Panels
451
Figure 13: Setup for generating panel tuned look-up table.
4.2 Computation of Content Change
Rate Coefficient (CCRC)
As discussed in section 3.1, significance of content
change analysis with reference to power saving and
quality assessment is very important parameter to
consider. The shortcomings in existing methods like
inter-frame brightness distortion, flickering effects,
clipping artifacts, color distortion is fixed by using
content change rate. Let’s consider video play use
case where content considered being high quality
and Content Change Rate (CCR) is min 30fps.
Existing work in this domain didn’t consider CCR.
Intensity and backlight change is applied on per
frame basis, causing inter-frame brightness
distortion. So in proposed technique CCR is
considered as one of decisive parameter for power
saving to eliminate these artifacts.
CCR is calculated based on number of frames
updated per unit time measured in frames per second.
CCRC calculation for different CCR is shown in
Table 3.
Table 3: CCRC computation based on CCR.
SL No. CCR CCRC
1 More than 60 0
2 45 to 60 1
3 30 to 44 2
4 20 to 29 3
5 10 to 19 4
6 8 to 9 5
7 6 to 7 6
8 4 to 5 7
9 2 to 3 8
10 1 or less 9
4.3 User Specified Aggressiveness
Proposed algorithm gives flexibility to user for
manually making algorithm more aggressive in
power saving, in case he is comfortable with slight
image degradation. This can be very useful when
user is running out of battery. Table 4 shows
representation of user aggressiveness based user
power level selection.
Table 4: Representation of user aggressiveness based on
power saving level.
SL No. User power saving level
1 low 0.5
2 mid 0.7
3 high 1
4.4 Computation of Effective
Aggressiveness
Effective Aggressiveness (

) including CCRC,
A
and CDC can be derived as (7)

= CCRC *max(CDC,A
))
(7)
Where
0 <=

<= 9
0 means no power saving
9 means highest power saving
Here max of CDCand A
is considered since
effective power saving should be max. Use of A

will be shown in next section.
4.5 Computation of New Intensity Level
(IL) and Backlight Level (BL)
The method of finding out new BL and IL using
A

and LUX_TABLE is described as below:-
Step 1: Read the current back-light level and find the
luminance (lux) at this back-light level (say BLc) by
indexing into the LUX_TABLE.
LUXc = LUX_TABLE (BLc, 0);
Step 2: Search for luminance (LUXc) in the
LUX_TABLE for all back-light level less than the
current backlight level(BLc) and higher intensity
levels as per below matching function to create
LUXMAP_ARRAY.
LUXMAP_ARRAY (BL, IL) = search (LUXc, threshold,
LUX_TABLE);
Where
For all BL < BLc and IL > 0
search(X, threshold, T(x,y)) is a function which
returns all sets of {x,y} for which the value
T(x, y) = X ± threshold.
This generates array of BL, IL pairs, producing
same amount of lux.
As explained in section 2, same lux can be produced
with different combinations of Backlight and
Intensity. Hence LUXMAP_ARRAY is array of BL
and IL pairs which produce same LUXc.
Display Color
analyzer
Black
Box
ICPRAM 2016 - International Conference on Pattern Recognition Applications and Methods
452
Step 3: Select the {BL, IL} pair from
LUXMAP_ARRAY by indexing the array at A
eff
, as
shown in below equation
.
[BLn, ILn] = LUXMAP_ARRAY [A
eff
]
Where
BLn is new backlight level to be applied
ILn is new Intensity level to be applied
New calculated BLn and ILn levels are applied to
backlight subsystem and display frame to get better
power saving without compromising on the quality.
5 EXPERMENTAL RESULTS
AND DISSCUSSION
5.1 Color Distribution Coefficient
Color Distribution Coefficient (CDC) was computed
for a set of 500 Simple UI and 500 Color rich
Table 5: SOM based CDC computation for different
images.
Image
SOM
Plot
CDM CDC Image
SOM
Plot
CDM CDC
1
81 0
6
27 0.46
2
65 0
7
39 0.22
3
93 0
8
44 0.12
4
82 0
9
23 0.54
5
67 0
10
37 0.26
Parameters selected for SOM are
Network Size = 6 x 6 Initial Neighbor Hood Radius = 6
Initial Learning Rate = .9 CDM Threshold (Th) = 50
Iteration Count = 32768
images. It was observed the computation correctly
classified nearly 93% of the test images. Also for
images with saturated colors, CDM was observed to
have higher value; hence this approach minimizes
the image modification to images having saturated
colors. Which can be justified since the image with
saturated colors have higher pixel values and so
stretch in SOM also becomes more with clusters
having lower pixel values.
Table 5 shows the value of CDM and CDC
computed using concept explained in section 3.2. It
can be observed that images having higher color
distribution have more stretch in SOM and hence
resulting in higher CDM value.
5.2 Content Change Rate Analysis
The proposed method was deployed on Galaxy
Mega Smartphone and Samsung Galaxy Tab AS and
usage data for multiple users for a period of 30 days
were recorded. Usage data is depicted in Table 6.
Table 6: Application wise display static percentage (> 2s)
for multiple users.
Application User 1 User 2 User 3 User 4
Call, contact
40% 35% 31% 29%
Facebook
22% 27% 23% 19%
WhatsApp
61% 13% 33% 27%
Launcher
19% 35% 23% 29%
Gallery
16% 8% 13% 17%
Average
31% 23.6% 24.6% 24.2%
Attempt was made to record the duration of static
content displayed while normal uses. So that
estimate for power saving can be done. It is seen in
Table 6; all popular applications have significant
static content.
Hence it can be inferred that there are many use
cases which have lower content change rate. So
there is a huge scope of power saving in algorithm,
since it is directly proportional to CCRC. The
algorithm saves power even in cases which have
high content change rate (low CCRC) but relatively
savings will be less.
5.3 Power Saving Analysis
Power saving for proposed technique is evaluated on
monsoon power monitor setup as shown in Figure
14. Power number measurements are done with and
without proposed technique for multiple use cases of
mobile device. Current consumption waveforms is
shown in Figure 15, showing around 80mA less
current consumption during static content use case
with the proposed solution.
Display Content Change Rate Analysis for Power Saving in Transmissive Panels
453
Figure 14: Power measurement setup.
After applying proposed technique on Galaxy
Mega 2 Smartphone, Samsung Galaxy Tab AS, test
shows positive results, with a current saving of
58mA to 80mA as shown in Figure 16. Power
savings was recorded for static use cases with
Simple UI image, to get an estimate of the maximum
savings.
Figure 15: Current consumption data with and without
proposed technique for browser use case.
Figure 16: Power saving data with and without proposed
technique.
Average power saving recorded with different
aggressiveness value on 4 devices is shown in
Figure 17. It can be observed that the amount of
power saving almost grows linearly with
Aggressiveness which is based on quality of image
and the rate at which the content is changing. Thus
least power saving achieved when, user display is
occupied with color rich image or a highly changing
content. When screen is having slow changing
content and simple UI, power saving is very high.
Figure 17: Average power saving for different
aggressiveness.
Figure 18, further shows the variation in
aggressiveness when user is using his phone in
launcher, contacts and video applications. It can be
seen when there is slow changing content (High
CCRC) and less color rich (high CDC) the
aggressiveness value is high and hence the power
savings.
Figure 18: Variation of

with respect to CCRC and
CDC.
As per the experimental results obtained in
Figure 16 for idle use cases, it can be inferred that
25% of the time device is in idle condition and
correspondingly 16% power saving can be achieved
in this condition. So effectively 4% power saving is
achieved throughout. It can be represented as
~50min increase in device usage time, if the device
is used for 20hrs. This increase in device usage time
is minimum power saving ensured by proposed
algorithm. Since calculation done above is only for
static content (i.e. CCRC = 9), but since algorithm is
operational and saves power in cases of slowly
changing content also it is bound to produce more
power saving.
Test Device
Power monitor
ICPRAM 2016 - International Conference on Pattern Recognition Applications and Methods
454
Further section discusses the quality assessment
of the proposed algorithm.
5.4 Quality Analysis
Quality analysis for different intensity and backlight
levels are evaluated based on Structural SIMilarity
(SSIM) (Wang et al., 2004). Figure 19 shows
experimental set-up, which consists of a mounted
Full HD Smartphone device to capture images of
test device with proposed algorithm running. Test
device is placed in closed container to avoid
surrounding light interference and test device is
connected to laptop via USB and controlled using
Android Debugger Bridge (ADB) command shell
interface.
SSIM analysis is done for different use cases at
different aggressiveness and analysis results are
depicted in Figure 20. Distortion due to proposed
algorithm is found to be negligible. Comparative
Figure 19: Experimental setup for Quality analysis.
Figure 20: Display image captured at different
aggressiveness.
analysis of image quality evaluated for different
aggressiveness with reference to luminance, contrast
and structure. Figure 21 shows quality analysis, the
error in contrast and luminance is very negligible
compared to structural index. Comparative study of
the image degradation results of presented algorithm
with the existing art (Hsiu et al., 2011) in this
domain reveals the proposed method achieves least
degradation.
Figure 21: Quality analysis for different aggressiveness.
Test application was made to invoke most used
applications, and flicker counts were recorded. It
was found that proposed method reduces flickering
artifacts by 88%. Results of same are shown in
Table 7.
Table 7: Comparison of Flicker counts of proposed and
conventional methods.
Use cases
Flickers recorded on Galaxy Tab AS
Proposed Conventional
Browser 1 6
Message 0 2
Launcher 0 3
Facebook 1 5
Fig. 22 shows the variation of PSNR along with
amount of power saving achieved. This is aligned
with our expectation as color rich image will have
more distortion and hence high PSNR and low
power saving. Also the more distortion we allow the
better power we can save due to lower level of
backlight. Average PSNR of TW-CES and (Kim et
al., 2010) was 26.32 dB and 27.11 dB respectively
and that of the proposed method was 29.05 dB
during high power saving. Hence the results show
the proposed method can extend the battery life by
saving power consumption at the same time
maintaining better user experience.
Black
Box
USB connector
for test device
USB connector
for camera
Display Content Change Rate Analysis for Power Saving in Transmissive Panels
455
Figure 22: PSNR measurement for different power saving.
6 CONCLUSIONS
The paper proposed a technique which is based on
Display content change analysis; Color distribution,
Backlight scaling and Intensity compensation to
achieve power saving in LCD class of displays. The
paper presented a 93% accurate image classification
method using Self Organizing Map to compute the
color distribution coefficient, which is used to
reduce chromatic distortion. It also showed
significance of low content change rate in mobile
devices and how they can be used to eliminate the
problems like inter-frame brightness distortion and
flickering by 88%, which are common problems for
all image luminosity compensated backlight scaling
methods. The proposed algorithm is prototyped on
Samsung Galaxy Tab AS and Galaxy Mega 2 and
power saving which is illustrated is encouraging.
The Structural similarity metric is used as image
distortion metric to evaluate the image distortion
which was found to be negligible and not
perceivable by human eye. The results show the
proposed method can extend the battery life
significantly by saving power consumption while
maintaining good user experience.
In future CDC computation algorithm will be
enhanced. Also contrast modification along with
intensity modification can improve the visual quality
more better which will be evaluated in the further
work of this research.
ACKNOWLEDGEMENTS
The authors would like to thank Mahammadrafi
Maniyar, Rajib Basu and Dibyadarshi Debadas from
System Software Team of Samsung R&D Institute
India Bangalore for providing support in this work.
REFERENCES
Carroll, A. and Heiser, G., 2010, June. An Analysis of
Power Consumption in a Smartphone. In USENIX
annual technical conference (Vol. 14).
Tsai, P.S., Liang, C.K., Huang, T.H. and Chen, H.H.,
2009. Image enhancement for backlight-scaled TFT-
LCD displays. Circuits and Systems for Video
Technology, IEEE Transactions on, 19(4), pp.574-
583.
Kerofsky, L. and Daly, S., 2006. Brightness preservation
for LCD backlight dimming. Journal of the Society for
Information Display, 14(12), pp.1111-1118.
Chen, M., Huajun, P.E.N.G., Qiu, G., Hung, C.K. and
Zhang, W., Hong Kong Applied Science and
Technology Research Institute Company Limited,
2013.LCD backlight dimming, LCD/image signal
compensation and method of controlling an LCD
display. U.S. Patent 8,514,166.
Choi, L., Shim, H. and Chang, N., 2002. Low-power color
TFT LCD display for hand-held embedded systems. In
Low Power Electronics and Design, 2002. ISLPED'02.
Proceedings of the 2002 International Symposium on
(pp. 112-117). IEEE.
Raman, N. and Hekstra, G.J., 2005. Content based contrast
enhancement for liquid crystal displays with backlight
modulation. Consumer Electronics, IEEE
Transactions on, 51(1), pp.18-21.
Cheng, W.C., Hou, Y. and Pedram, M., 2004, February.
Power minimization in a backlit TFT-LCD display by
concurrent brightness and contrast scaling. In
Proceedings of the conference on Design, automation
and test in Europe-Volume 1 (p. 10252). IEEE
Computer Society.
Hsiu, P.C., Lin, C.H. and Hsieh, C.K., 2011, August.
Dynamic backlight scaling optimization for mobile
streaming applications. In Proceedings of the 17th
IEEE/ACM international symposium on low-power
electronics and design (pp. 309-314). IEEE Press.
Schatzmann, J. and Ghanem, M., 2003. Using self-
organizing maps to visualize clusters and trends in
multidimensional datasets. Department of Computing
Data Mining Group, Imperial College, London, p.132.
Kohonen, T., 1990. The self-organizing map. Proceedings
of the IEEE,78(9), pp.1464-1480.
Wang, Z., Bovik, A.C., Sheikh, H.R. and Simoncelli, E.P.,
2004. Image quality assessment: from error visibility
to structural similarity. Image Processing, IEEE
Transactions on, 13(4), pp.600-612.
Kim, T.H., Choi, K.S. and Ko, S.J., 2010. Backlight power
reduction using efficient image compensation for
mobile devices. Consumer Electronics, IEEE
Transactions on, 56(3), pp.1972-1978.
ICPRAM 2016 - International Conference on Pattern Recognition Applications and Methods
456