Computational Water, Energy, and Environmental Engineering, 2013, 2, 26-30
doi:10.4236/cweee.2013.23B005 Published Online July 2013 (http://www.scirp.org/journal/cweee)
Digital Interactive Kanban Advertisement System
Using Face Recognition Methodology
Feng-Yi Cheng, Chu-Ja Chang, Gwo-Jia Jong
Department of Electronics Engineering, National Kaohsiung University of Applied Sciences, Kaohsiung.
Received May, 2013
ABSTRACT
Most of advertisement systems are presently still launch the publicity content by the static words and pictures. Recen tly,
this static advertisement model will not be able to attract people’s attentio n more and more. Moreover, the static infor-
mation content of advertisement system is limited because of the layout shown size. It can not also fully demonstrate
the information content of advertisement system. In this paper, we develop a digital interactive kanban advertisement
system using face recognition methodology to solve these problems. The system captures the person’s face through the
camera. The digital advertisement con tent size is relevant by the person and camera observation locations. In this paper,
we adopt the Adaboost algorithm to judge people face, and the system only need to grab the position of the face. The
system doesn’t built expensive and co mplex equipment to reduce the system cost and enhance the system performance.
This system can also achieve the same similar digital interactive advertising effectiveness.
Keywords: Face Recognition; Kanban Advertisement; Adaboost; Interactive
1. Introduction
There are various kinds of advertising media now. For
example: televisions, broadcastings, magazines, news-
papers, outdoor advertising and transit advertising…etc.
Beside above media, the appearance of the web advertis-
ing makes advertising having more and more develop-
ment space. An advantage of web advertising is that
consumers can select advertising freely and obtain mes-
sages advertising transmit immediately. Advertising can
interact with consumers and transmit feedback instantly,
but traditional advertising can’t [1,2]. In this paper, in-
teractive multimedia as the theme to explore the visual
interface design and interactive multimedia development,
according to the popular trend of today’s interactive mul-
timedia authoring the common interactive multimedia
such as: interactive web pages, DVD movies menu, web
advertising, teaching CD-ROM…etc., because of the
rapid growth of technology and the technology matures,
the traditional static advertising, web, gradually interac-
tive multimedia replaced [3].
In recent years, face detection technology tends to
mature, the technology has been widely used in cameras,
computer identification system and interactive advertis-
ing. In [4], they research proposes a system that engages
audience to the advertisement through interactive appli-
cations and provides data to the advertiser/producer
about their audience, but we think that is too complex.
Therefore, we propose an interactive system, using text
and pictures to do the interactive display, it can allow
users to quickly understand the contents of advertise-
ments.
2. Interactivity Kanban Advertisement
The Figure 1 is the configuration of interactive kanban
advertisement system. At first, we capture the camera
image, then color segmentatio n extracts the skin color of
a face from a cluttered image; then, binary imaging fur-
ther forms a more complete region. Next, morphological
erosion eliminates some of the small spots in a tested
image. Contrary to erosion, dilation enlarges and con-
nects a small and disconnected, but marked, facial region.
Subsequently, connected component labeling is em-
ployed to mark multiple faces in the image. Finally, an
area threshold and an aspect ratio are used to validate th e
corrected facial region. After then we use the Adaboost
algorithm [7-10] to make face recognition. At last, we
can judge by the distance between the captured face and
advertising, when the person is closer and closer, the
words will accord to the distance for scaling to achieve
Capture
Image Face Recognition
Interactive
Advertisements
Figure 1. Interactivity kanban ad syste m bloc k diagram.

F.-Y. CHENG ET AL. 27
the purpose of attracting users.
The Figure 2 is our hardware configuration of capture
image, the camera mounts on the Kanban Advertisement
above.
3. Face Recognition
The Figure 3 is our face recognition system, the face
recognition has three processing steps: Skin color detec-
tion, Detect face region, facial feature points and finally,
through Adaboost screening the most right face of people.
This part is mainly to capture each person’s face form the
image.
3.1. Skin Color Detection
The color segmentation is an important pre-processing
step in the face recognition methods. We are used HIS
method [5] to detection face skin color. Frist, we trans-
form the image of RGB three color changes to H, S. It
can be written as
,
360
if BG
Hif BG






(1)
where


11/2
2
1()
2
=cos RG RB
RG RBGB



 





 








(2)


3
1min(,SR
RGB
 

,)GB (3)
Then we can follow the rule to find skin color fc:
camera
Kanban
Advertisement
Figure 2. Interactivity kanban ad syste m configur ation.
Figure 3. Face recognition block diagram.
1, 832 30163
0,
c
if HandS
fotherwise
 


 (4)
After the skin color detection, we only see the portion
of skin color, as shown in Figure 4.
3.2. Face Region Detection Method
After the Skin color detection, we change the RBG to
binary used thresholding. Then, we consulted the paper’s
method [6] to the binary image is sub-divided into blocks.
Then, the total skin area within a b lock is computed, and
if this is greater than or equal to 40% of the block area,
the block label is assigned to be skin. A connected region
step is then performed by examining the 8 neighbour-
hood connectivity among the blocks to create a set of
candidate regions. Face regions are selected amongst the
candidate regions as the regions having an aspect ratio
corresponding to the 1.2 - 2.0 ratio, than use the rate
value of the image to d efine the threshold. It is showed in
Figure 5.
3.3. Capture Facial Feature Points
3.3.1. Eyes Detection
It is obvious that eyes are non-skin color regions, the Cr
and Cb component of eyes and skin contains bigger dif-
ference in the YCrCb space, and the Cb is higher than the
Cr in the eyes region. It can detection location and size
by above information.
3.3.2. M ouths Detecti on
Mouth is also non-skin color region, in the mouth the Cr
is much higher and the Cb much lower. Increasing the
difference between the Cb and Cr can accurately detected
size and location.
3.4. Adaboost Algorithm
Machine learning algorithm is flourishing in recent year,
widely used at various levels. Face Detection this issue in
order to obtain better characteristics also introduces a
machine learning concepts, these studies are a break-
through in the past to the face detection frame, most no-
tably the 2004 study is presented using the integral image
Viola for the characteristic value of the AdaBoost face
detection method. AdaBoost is an algorithm for con-
structing a “strong” classifier as linear combination of
“weak” classifiers [7-10].
3.4.1. Haar-Like Features and Integral Image
A set of Haar-like features, used as the input features to
the cascaded classifiers, are shown in Figure 6. In our
work, Haar-like features consideration is using integral
image to improve computation efficiency.

F.-Y. CHENG ET AL. 28
(a) (b)
Figure 4. (a) Original image; (b) Skin color detection.
(a) (b)
Figure 5. (a) Morphological process; (b) Face region detec-
tion.
Figure 6. The Haar-like features for AdaBoost algorithm.
The Haar-like features that shown in Figure 6. It used
in our face detection system. The features can be rapidly
computed at different scales by introducing “Integral
Image”.
3.4.2. Ad aB oost Algorithm
In fact AdaBoost is a classification of concepts, for ex-
ample, In order we pick a better than normal a little bit
(>= 50%) of the algorithm, it can again and again use
update weighting approach to reduce error rate, The pro-
cess is as follows
1) Input M sample of the target image and N sample
not of the target image, and I search the number of fea-
tures.
2) Initialize weights
Target image samples weights
1
2
m
WP
M
 (5)
Non-target image samples weights
1
2
n
WN N  (6)
3) For each feature j, train a weak classifier T, and
evaluate its error E with respect to W
11
(1 )
MN
mm n
mn
EWPT WNT

n

 

(7)
In this derivation when T = 1 consistent with the image
features, T = 0 does n ot meet.
4) Using step 3. Add the choose features to the stage
and determine the corresponding weights
1
log
iE
WE






(8)
5) For step 3. Searching the better features to updates
image sample weights. Updated target image sample
weights.
1
1
E
mm
E
WP WPE






(9)
Updated non-target image samples weights.
1
1
E
nn
E
WN WNE






(10)
6) Normalize the weights
1
m
mM
m
m
WP
WP
WP









(11)
1
n
nn
n
n
WN
WN
WN









(12)
7) Check whether the number of the current search
features to meet the demand, if the lack of jump back to
step 3, otherwise the e nd of.
4. Interactive Advertisement System
The Figure 7 is our interactive advertisement system
flow process. At first, we capture the camera’s image,
Figure 7. Interactive ad system flow process.

F.-Y. CHENG ET AL. 29
then thought the image to do face recognition processing,
to get the information of people who watch the Adver-
tisement, and then the system further determine whether
capture face or not, in other words, to determine if some-
one is watch the kanban advertisement system, after then
to detection the distance between the person’s face and
advertisement, when people are approached the adver-
tisement, the advertisement will also show more message
telling the people, let people can learn more about the
details of the advertisement, to increase the impression of
people watch advertisement.
5. Result and Discuss
We use a 20 million pixels webcam and a 36-inch TV to
achieve the Interactive Kanban Advertisement System.
The webcam is set in place of 130cm high and angle of
90 degrees. The Figure 8 shows the information that the
relationship between pixel size of face and the distance
from person to camera. When the system captures the
person’s face, we can use the pixel size of face to deter-
mine where the person.
The Figure 9 is the user interface of program, this pro-
gram of interface can divided into two parts, the camera
of image is on the left, the interactive kanban advertise-
ment is on th e righ t. If so meon e walks p ast in fron t of th e
kanban advertisement, the system will catch person’s
face, and the kanban advertisement also shows some
words to attract people’s attention, it is like Figure 9.
When person is closer and closer, the system will cal-
culate the face of pixels to detect the distance, if people
rely on close enough, the system will change the adver-
Figure 8. Graph of ratio between person and camera.
Figure 9. The user interface of program.
Figure 10. It change the ad content when people rely on
close enough.
tising content, display more information attract people to
continue to watch, it is like Figure 10.
6. Conclusions
In this paper, we propose an Adaboost algorithm ap-
proach to the face recognition for applications of interac-
tive advertisement system. Although it has a lot of people
to research for the subject so far, however, the proposed
approach are more complex, build may more cost or need
to take the time to interact with advertisement, caused
people inconvenience. Our system only through the dis-
tance between the face and advertising to interact,
through words and pictures to attract people, reduce the
complexity of the system also allows people quickly to
understand more information of advertisements.
REFERENCES
[1] J. H. Cho, Y. J. Sah and J. Ryu, “A New Content-related
Advertising Model for Interactive Television,” Broadband
Multimedia Systems and Broadcasting 2008, March 31
2008-April 2 2008, pp. 1-9.
[2] M.-H. Hsieh, D.-L. Yang and J.-Y. Dai, “A Face Recog-
nition System Prototype to Evaluate the Effectiveness of
Digital Advertisement,” 2010 Conference on Computer
Vision, Image Processing and Information Technology,
2010-06. Zhongli, Taiwan, pp. 283-289.
[3] J. Kim and S. Kang, “An Ontology-Based Personalized
Target Advertisement System on Interactive TV,” Con-
sumer Electronics (ICCE), 2011 IEEE International
Conference, 9-12 Jan. 2011，pp. 895 - 896.
[4] M. Taspinar, A. T. Naskali, M. Kurt and G. Eren, “The
Importance of Customized Advertisement Delivery Using
3D Tracking and Facial Recognition,” in Proc. The Sec-
ond International Conference on Digital Information and
Communication Technology and its Applications (DIC-
TAP), 2012, pp. 526-530.
[5] S. Guerfi, J.-P. Gambotto and S. Lelandais, “Implementa-
tion of the Watershed Method in the HSI Color Space for
the Face Extraction,” Advanced Video and Signal Based
Surveillance, Sept. 2005, pp. 282-286.
[6] M. Rahman and N. Kehtarnava z, “Real-T imeFace-Priorit
y Auto Focus for Digital and Cell-Phone Cameras,” IEEE
Transactions on Consumer Electronics, Vol. 54, No. 4,
2008, pp. 1506-1513.doi:10.1109/TCE.2008.4711194

F.-Y. CHENG ET AL. 30
[7] J. X. Ruan and J. X. Yin “Multi-Pose Face Detection
Using Facial Features and AdaBoost Algorithm,” Second
International Workshop on Computer Science and Engi-
neering, 2009, pp. 31-34.
[8] Y.-W. Wu and X.-Y. Ai, “Face Detection in Color Im-
ages Using AdaBoost Algorithm Based on Skin Color
Information,” Workshop on Knowledge Discovery and
Data Mining, 2008, pp. 339-342.
[9] S. A. Inalou and S. Kasaei “AdaBoost-Based Face Detec-
tion in Color Images with Low False Alarm,” Second In-
ternational Conference on Computer Modeling and
Simulation, 2010, pp.101-111.
doi:10.1109/ICCMS.2010.287
[10] Y. C. Xing, Z. Z. Wang and W. P. Qiang, “Face Tracking
Based Advertisement Effect Evaluation,” Image and Sig-
nal Processing, 2009. CISP '09. 2nd, 2009, pp. 1-4.