www.ierjournal.

org

International Engineering Research Journal (IERJ) Volume 1 Issue 4 Page 122-125, 2015, ISSN 2395-1621

A Real Time Virtual Fitting Room

Application

ISSN 2395-1621

Nutan Kumari #1, Surabhi Bankar #2

1

[email protected]
surbhi.bankar @gmail.com

Guided by
Prof. Prutha Ardhapure.
#12
Dept. of Computer Engineering, G.S.M.C.O.E.
University of Pune, Maharashtra, India

ABSTRACT

ARTICLE INFO

It is really time consuming activity to try clothes in clothing stores. In this paper we
present a Real Time Virtual Fitting Room Application which allows customers to choose
clothes in virtual environment. We provide an easier way to try & choose clothes
virtually. Our application allows user to try clothes of various sizes, different styles &
proper fittings in virtual environment rather than physically trying them. For this we
use image of customer, models of different coordination and skin color of that customer.
The presented application is used to improve accessibility of trying clothes and
maximizing the time efficiency by introducing a virtual fitting room environment.

Article History
Received : 17th April, 2015
Received in revised form :
22nd April, 2015
Accepted : 25th April, 2015
Published online :
29th April, 2015

Keywords Virtual Fitting Room (VFR), virtual space, 2D-modelKeywords Social

Wireless Network, Cooperative caching, selfish user, Content provisioning, ad hoc
networks.

I.

INTRODUCTION

There is substantial, loss of time in don and doffing of

clothes in stores which is one of the common tasks which
requires most of the time. Generally more waiting time have
to be considered, for example, when looking into the mirrors
of changing rooms. Even, more time is wasted for don and
doffing, and also most buyers doesn't like to purchase
garments online directly or they are not satisfied with their
online shopping experience.
Clothing descriptors of anatomical types are more mixed
and less experimental, e.g. "outsize", "flat-chested" or,
"pear-shaped". The available measurements about body
shapes are largely unreliable and most clothing is made to
fit a small number of stands, which are assumed to represent
average sizes. The explanation is extraordinary custom
and routine, with little regularity in the market place and
continuing customer needs about fittings. Shape analysis
allows the perfect and accurate averaging of body shapes
which fall into a particular size category, allowing to
improved mannequins (real and virtual) to be made.
The techniques discussed in this paper can enhance the
shopping involvement. We are proposing a Virtual Fitting
Room (VFR) system, which gives a answer for the above
conditions. We are developing function that is based on
software which helps in representing output from the
structure, extracted from image (taken from camera).When

2015, IERJ All Rights Reserved

the person is standing in front of the camera, he will be able

to select clothes whatever he wants. We can also expand our
system in future to recommend some clothes depending on
his skin color which will suit on that specific person. We are
virtually assigning cover of the selected garment with the
image recorded by the camera.
ORGANIZATION
The paper is organized as follows: Existing System is
presented in Section II.We present our scheme in Section III.
The conclusion in Section IV .Future work in Section V
II. EXISTING SYSTEM
A system employs one Kinect sensor and one HighDefinition (HD) Camera. System has been deployed since
April 2012 in one of Singapores largest shopping centers.
In order to get a convincing virtual try-on experience for the
customer user, the Kinect sensors are being used. These
sensors are built by Microsoft and are very expensive. HD
camera is used to replace the role of Kinects built-in RGB
camera included for HD recording. This necessitates a
calibration process between the HD camera and the Kinect
depth camera in order to map the 3D clothes seamlessly to
the HD video recording of the customers.
Virtual try-on system consists of a vertical TV screen, a
Microsoft Kinect sensor, an HD camera, and a desktop
computer. Figure gives idea of the front view of the

Page 1

www.ierjournal.org

International Engineering Research Journal (IERJ) Volume 1 Issue 4 Page 122-125, 2015, ISSN 2395-1621

Interactive Mirror together with the Kinect and HD camera.

The Kinect sensor is an input device marketed by Microsoft,
and designed as a gaming interface for Xbox 360 consoles
and PCs. It includes a depth camera, an RGB camera, and
microphone arrays. Depth and camera have horizontal and
vertical viewing ranges of 57.5 and 43.5 degrees.
Kinect can also angle up and down within -27 to +27
degrees. The territory of the depth camera for normal mode
is [0.8_4] m and for near mode it is [0.4_3] min. The HD
camera supports a full resolution of 2080 _ 1552, from
which Virtual Try-on using Kinect and HD camera.

motion tracking component implemented on two publicly

available Kinect SDKs, as will be discussed in Section 4.
The user interacts with the Interactive Mirror with her right
hand to control the User Interface (UI) and select clothing
items.
When the users feet are not in the field of view of Kinect.
The tilting angle of the Kinect sensor, the depth of the neck
joint, and the offset of the neck joint with respect to the
center point of the depth image can jointly determine the
physical height of the neck joint in the world space.
III. PROPOSED SYSTEM
Proposed VFR is software based and designed to be
universally compatible as long as the device has a camera.
The use of web camera is a cheaper alternative to Kinect
sensors. It does not require extra hardware support. The
users can use the proposed system from their home itself. It
provides real time access. Compared to other existing VFR
systems, key difference is the proprietary hardware
components or peripherals. The system makes the use of
web cam to detect the human body. The body is then
divided into upper body and lower body. Resizing of the
images is done to superimpose the cloth image on the
human body. This is cheaper version of the existing system
which uses lot of hardware and cannot be used at home

Fig. 1 The front view of the Interactive Mirror with Kinect and HD
camera placed on top.

Fig. 3 Proposed View Of System

OVERALL FLOW OF PROPOSED SYSTEM

The Fig. 4 .shows how proposed system will performs the
operations.
Fig. 2 Shoulder height estimation

Fig. 2 illustrates the major software components of the

virtual try-on system. During the offline preprocessing stage,
we need to calibrate the Kinect and HD cameras, and create
3D clothes and accessories. These two components will be
discussed in more details in next sections respectively.
During the online virtual try-on, we first detect the nearest
person among the people in the area of interest. This person
will then become the subject of interest to be tracked by the

2015, IERJ All Rights Reserved

Page 2

www.ierjournal.org

International Engineering Research Journal (IERJ) Volume 1 Issue 4 Page 122-125, 2015, ISSN 2395-1621

Fig. 5. Background removal. Depth image (left), color

image (middle), extracted user image (right)
A threshold is applied to the color components of the image
within the following ranges:

77 < Cb < 127

177 < Cr < 173

Y < 70
Fig. 4 System Architecture

3.1 User Extraction

Derivation of user allows us to create an
augmented reality environment by isolating the user area
from the image and superimposing it onto a virtual
environment in the user interface. Additionally, it is very
useful way to know the region of interest that is also used
for finding skin color which is explained in further section
The camera gives the image. When the device is used state,
image is divided in order to separate background from the
user. The background is discarded by blending the RGBA
image with the divided image for each pixel by setting the
alpha channel to zero if the pixel does not belong to the user.

Since we have the extracted user image as a region of

interest the threshold is given only on the pixels that belongs
to the user. Thus, the areas on the background which may be
similar with the skin colour are not processed. The skin
colour segmentation is illustrated in Figure 6.

3.2 Skin Segmentation

Since the model is superimposed on the upper layer, the user
always stops behind the model which does not allow some
capable actions of the user such as folding arms or holding
hands in front of the t-shirt. In order to clear up that problem
skin colour areas are searched and brought to the front layer.
HSV and YCbCr color spaces are generally used for skin
colour dividing. In this work we used YCbCr colour space
and the RGB images are transformed into YCbCr colour
space by following equations:

Fig. 6. Skin color segmentation. User image (left), segmented image

(right)

Cb =128-0.169R-0.33G+0.5B

3.3 HAAR Classifier

Y = 0.229R+0.587G+0.114B

The core basis for Haar classifier object detection

is the features which are similar to Haar. These features,
other than using the force values of a pixel, use the different
contrast values between neigh boring rectangular groups of
pixels. The contrast differences between the pixel groups are
being used to find whether relative light and dark areas are
present. Two or three neighbouring groups with a relative
contrast variance form a Haar-like feature. Haar-like
features, as given in Figure are used to find an image . Haar

Cr = 128+0.5R-5.419G

2015, IERJ All Rights Reserved

Page 3

www.ierjournal.org

International Engineering Research Journal (IERJ) Volume 1 Issue 4 Page 122-125, 2015, ISSN 2395-1621

features can simply be scaled by increasing or decreasing

the size of the pixel group being determined. This uses
features to detect objects of different sizes.
3.4 Integral Image
The simple rectangular features of an image are
calculated using an intermediate representation of an image,
called the integral image . Array which is also known as
integral image which containing the sums of the pixels
intensity values located directly to the left of a pixel and
directly above the pixel at location (x, y) inclusive . So if
A[x,y] is the original image and AI[x,y] is the essential
image then the integral image is computed as shown in
equation and illustrated in Figure 7

IV. CONCLUSION
After an introduction, the related work was presented;
starting with Cloth selection and virtual try-on, cloth
recommendation system is also present there. Afterward a
convenient look on the technologies and frameworks that
are being used for the creation, like Haar classifier
algorithm, of the Measurement of the tailoring and Virtual
Try-on was taken. After this the different conditions of the
design process up to the construction of the garment models
were highlighted.
In the last section the tests was carried out, also
discussion on outcome, the appearance and the interaction
with the Tailoring Measurement and Virtual Try-on. Overall,
the proposed Virtual Dressing Room is a good quick fix for
easy and accurate try-on of garment. In this system
compared to other technologies like augmented reality
markers or real-time motion capturing techniques no
expensive configurations and time-consuming build-ups are
required. After this point of view it is an excellent addition
for a cloth store.
Beyond that a simple setup of the system can also be
assembled at home since the minimum requirements are a
computer with a screen and a Camera. This can also
outcome in an additional feature for a web shop, for instance.
It would allow a virtual try-on of clothes before people are
shopping it online, looking closer at the garment and even
conveying the actual behavior of the real cloth. This gives a
huge benefit over the common web shopping experience.
V. FUTURE WORK

Fig. 7 Summoned are of integral image

3.5 Classifiers Cascaded

Although calculating a feature is extremely efficient and
fast, computing all 180,000 features present within a 24
24 sub-image is impractical [Viola 2001, Wilson 2005].
Fortunately, only a tiny fraction of those features are needed
to determine if a sub-image potentially contains the
expected object. In order to discard as many sub-images as
possible, only a few of the features that characterize an
object are used when analysing sub-images. The goal is to
discard a substantial amount, around 50%, of the subimages that do not contain the object.
Haar classifiers continue, increasing the number of
features used to analyze the sub-image at each stage. The
cascading of the classifiers uses only the sub-images with
the highest possibility to be analyzed for all Haar-features
that discuss an object. It also uses one to find the accuracy
of a classifier. One can maximize both the false alarm rate
and positive hit rate by minimizing the number of stages.
The inverse of this is also true. Viola and Jones were able
to accomplish a 95% accuracy rate for the detection of a
human face using only 200 simple features. Using a 2 GHz
computer, a Haar classifier cascade could detect human
faces at a rate of at least five frames per second

2015, IERJ All Rights Reserved

Based on a acceptable duty and pricing case, a debatable

model for the content providers price computing is created.
A cooperative caching method, Split Cache, is desired,
numerically analysed, and theoretically confirmed to give
best possible object placement for systems with
homogenous content requirements. A benefit-based
approach, Distributed Benefit, is used to minimize the
provisioning price in heterogeneous networks consisting of
nodes with different content request rates and patter.
VI. CONCLUSIONS
The main aim to this work was to develop a cooperative
caching strategy for provisioning cost minimization in
Social Wireless Networks. The key improvisation is to
display that the best cooperative caching for provisioning
cost reduction in networks with homogeneous content
demands requires an optimal split between object
duplication and individuality. In addition to that, we
experimentally (using simulation) and analytically evaluated
the algorithms performance in the presence of user
selfishness. It was well known that selfishness increases
user rebate whenever there are selfish nodes in a Social
Wireless Network having less than a critical number. It was
given that with heterogeneous requests, a profit based
heuristics methods gives better achievements compared to
split cache which is desired mainly for homogeneous
request. Ongoing work on this topic uses the development
of an efficient algorithm for the heterogeneous call scenario,
with aim of bridging the achievements gap between the
profits Based heuristics and the centralized greedy method

Page 4

www.ierjournal.org

International Engineering Research Journal (IERJ) Volume 1 Issue 4 Page 122-125, 2015, ISSN 2395-1621

which was convinced to be optimal Removal of the nocollusion belief for user selfishness is also being worked on.
Future work on this topic have the development of an
efficient algorithm for the heterogeneous demand scenario,
with a goal of bridging the performance gap between the
Benefit Based heuristics and the centralized greedy
mechanism which was proven to be optimal in can be
removal of the no-collusion assumption for user selfishness
is also being worked on.

REFERENCES
[1]

M. Zhao, L. Mason, and W. Wang, Empirical Study

on Human Mobility for Mobile Wireless Networks,
Proc. IEEE Military Comm. Conf. (MILCOM), 2008.

[2] Cambridge Trace File, Human Interaction Study,

http://www.crawdad.org/download/cambridge/haggle/
Exp6.tar.gz,2012.
[3] E. Cohen, B. Krishnamurthy, and J. Rexford,
Evaluating Server-\ Assisted Cache Replacement in
the Web, Proc. Sixth Ann.European Symp.
Algorithms, pp. 307-319, 1998.

[13] H.K. Kuhn, The Hungarian Method for the

Assignment Problem, Naval Research Logistics, vol
52, no. 1, pp. 7-21, 2005.
[14] N. Laoutaris et al., Mistreatment in Distributed
Caching: Causes and Implications, Proc. IEEE
INFOCOM, 2006.
[15] B. Chun et al., Selfish Caching in Distributed
Systems: A Game-Theoretic Analysis, Proc. 23th
ACM Symp. Principles of Distributed Computing,
2004.
[16] M. Goemans, L. Li, and M. Thottan, Market Sharing
Games Applied to Content Distribution in Ad Hoc
Networks, IEEE J. Selected Areas in Comm., vol. 24,
no. 5, pp. 1020-1033, May 2006.
[17]

National Laboratory of Applied Network Research,

Sanitized Access Log, ftp://ircache.nlanr.net/Traces,
July 1997. [24] NASA Kennedy Space Center, WWW
Server
Access
Log,
ftp://ita.ee.lbl.gov/traces/NASA_access_log_Jul95.gz, 2012.

[4] S. Banerjee and S. Karforma, A Prototype Design for

DRM Based Credit Card Transaction in ECommerce, Ubiquity, vol. 2008,
[5] L. Breslau, P. Cao, L. Fan, and S. Shenker, Web
Caching and Zipf-Like Distributions: Evidence and
Implications, Proc. IEEEINFOCOM, 1999.
[6] C. Perkins and E. Royer, Ad-Hoc On-Demand
Distance Vector Routing, Proc. IEEE Second
Workshop Mobile Systems and Applications,
[7] S. Podlipnig and L. Boszormenyi, A Survey of Web
Cache Replacement Strategies, ACM Computing
Surveys, vol. 35, pp. 374-398, 2003.
[8]

A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass,

and J. Scott,
Impact of Human Mobility on Opportunistic
Forwarding Algorithms, IEEE Trans. Mobile
Computing, vol. 6, no. 6,pp. 606-620, June 2007.

[9]

BU-Web-Client - Six Months of Web Client Traces,

http://www.cs.bu.edu/techreports/1999-011-usertrace98.gz, 2012.

[10] A. Wolman, M. Voelker, A. Karlin, and H. Levy, On

the Scale and Performance of Cooperative Web
Caching, Proc. 17th ACM Symp.
Operating Systems Principles, pp. 16-31, 1999.
[11] S. Dykes and K. Robbins, A Viability Analysis of
Cooperative Proxy Caching, Proc. IEEE INFOCOM,
2001.
[12] M. Korupolu and M. Dahlin, Coordinated Placement
and Replacement for Large-Scale Distributed Caches,

2015, IERJ All Rights Reserved

Page 5