International Journal of Scientific & Engineering Research, Volume 7, Issue 4, April-2016 82

ISSN 2229-5518

Robust Algorithm for Developing Barcode

Recognition System using Web-cam
Sampada S. Upasani, Adarsh N. Khandate, Ankita U. Nikhare, Rupali A. Mange, R.V.Tornekar

Abstract— This paper presents an algorithm to develop a Supermarket billing system, which recognizes the barcode using image
processing. MATLAB software is used to test the algorithm. There are various types of barcodes in use today. These include Code 128,
Code 39, EAN 13, etc. Our algorithm decodes EAN 13 barcode that is used universally. In this algorithm, the database of products is
created in an excel sheet. Images of products are taken using USB webcam, which is to be installed manually in MATLAB Image
Acquisition Toolbox. Preprocessing is done that includes processes such as binarization; spatial filtering and edge detection on these
images to extract the barcode region and the barcode number is decoded using the decoding algorithm. This barcode number is further
compared with the database, and the corresponding information of the scanned product is displayed in a different excel sheet. A final bill of
all scanned products is created after scanning of all products. Proposed System provides a convenient method of extracting information
from the barcode at a lesser cost as compared to typical electronic barcode scanners. This system can be used by businesses requiring
accurate results at low cost.

Index Terms—Barcodes, Barcode localization, Contrast enhancement,EAN-13,Edge detection, Threshold, UPC-A

————————————————————
1 INTRODUCTION
Barcode is a visual depiction of information in the form of
bars and spaces on a surface. The bars and spaces can be of

IJSER
different widths and consists of characters, numbers and
symbols such as colon, dot, and others. Set of various
combinations of these alphanumeric characters is used to
depict information. There are different types of barcodes in
use today. These include Code 128, Code 39 [1], and EAN
13.The barcodes are read using scanners that use laser
beams or cameras. In the supermarket, there are many
hand-held scanners [2] that automatically scan the barcode
on the merchandises. This paper discusses an algorithm for
barcode recognition system by using a web camera or
digital camera for capturing barcode image and then
display the barcode information for the user. Nowadays
most of the barcode scanners are using infrared to scan a Fig. 1.Conceptual analysis diagram
barcode. These types of scanners may lead to the cost issues
as those scanners are expensive and unaffordable to the This preprocessed image is then given to the decoding
common user. These cost issues can be solved by using the algorithm to extract the barcode number. Comparison of
camera-based system for reading barcodes. this number with the database extracts the product
It involves two steps for automatic scanning of all the information.
barcodes:
1. To find the position of a barcode
2. To decode the barcode.
2 EAN-13 OVERVIEW
Unfortunately, images taken by a web camera or digital An EAN-13 barcode was originally called European Article
camera [3] are often of low quality. In this system, the Number but now renamed as International Article
proposed algorithm detects the barcode region using Number, although the abbreviation EAN has been retained.
gradient method and this region is then cropped. Image- It is a 13 digit (12 data and 1 check) bar coding standard. It
preprocessing module transforms the cropped barcode is a superset of the original 12-digit Universal Product Code
image into a preprocessed grayscale image, then reduces (UPC) [4] system developed in the United States. Fig. 2
noise, and enhances the contrast between bars and spaces shows the EAN-13 barcode. As EAN-13 is considered as a
in this preprocessed image. superset of UPC-A, any software or hardware that is
capable of reading an EAN-13 symbol can also read a UPC-
A symbol.

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 4, April-2016 83
ISSN 2229-5518
TABLE 2
ENCODING OF L,G,R CODES

Fig. 2.EAN-13 Barcode

The following process is followed for binary encoding of

data digits into EAN-13 barcode: Firstly, the digits are split
into three groups such that the first digit, the first group of
6 and the last group of 6, form three distinct set of numbers.
The first group of six is encoded using a scheme where each
digit has two possible encodings, one that has even parity
and one that has odd parity. The first digit is encoded by
selecting a pattern of choices between these two encodings
3 PREVIOUS WORK
for the next six digits, as shown in Table 1 and Table 2. (As
opposed to other digits, the first digit is not represented by Barcode reading has been studied for decades. And after

IJSER
a pattern of bars). Every digit in the last group of six digits continuous improvement, it is now represented by a well-
is encoded using a single set of bar-patterns that are the established industrial standard. Until recently, barcode
same patterns that are used for UPC. If the first digit is reading was performed almost exclusively with dedicated
zero, all digits in the first group of six are encoded using hardware. Even though there is growth in the use of
the patterns that are used for UPC; therefore, a UPC camera-based readers, most of the challenges posed by this
barcode is also considered as an EAN-13 barcode with just new method are yet to be solved. Commercial scanners that
the first digit set to zero. are used in supermarkets shine a stripe of pulsed light on
the barcode and record the intensity of its reflection. As it
makes use of active illuminations, it makes barcodes
TABLE 1 virtually insensitive to changes of ambient illumination.
ENCODING OF EAN-13 Additionally, their design often requires the barcodes to be
relatively close to the scanner to scan it successfully. In
general, these dedicated devices produce a high-quality
signal that allows for robust barcode reading. On the other
hand, reading barcodes with cameras present new
challenges. The major issue faced is with the poor quality of
the image due to noise or low contrast.
Orazio Gallo and Roberto Manduchi[5] proposed a new
approach to barcode decoding that bypasses binarization.
Their technique relies on deformable templates and exploits
all of the gray-level information of each pixel. They have
used UPC-A symbol of barcode and presented a new
algorithm for barcode decoding (localization and reading)
that can deal with blurred, noisy and low-resolution
images.
Luping Fang and Chao Xie [6]present an algorithm, which
can detect barcode region even in a complex background
with the help of region-based image analysis. Being
different from traditional region-based image analysis, the
algorithm is processed with an integral image. It can deal
with images that are blurred, contain obliquely positioned
barcode regions and are shot under different illumination.
Xianyong Fang[1]has proposed a fast and robust
recognition method for noisy Code 39 barcode. The
IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 4, April-2016 84
ISSN 2229-5518
proposed method follows two steps: search and decoding. The size of the filter was selected based on the range of the
In the first step, all asterisks in the image are found with size of the input barcode images and the minimum size of
evenly defined scan lines and then those lines with the the barcode readable by our method. Note that block
same directions are matched together to get a valid barcode filtering can be implemented efficiently so that only a few
region. In the second step, a local de-noise method is operations per pixel are required. In the end, we binarize
applied for eliminating noise in the barcode region. Further, Is(n) with a single threshold that is selected using the
a middle band filter is used for decoding the barcode. Their method proposed by Otsu [7], and the output image is as
method also has certain shortcomings. For images whose shown in Fig. 5. This binarization is only used to localize
resolution are quite different from 2400x3500, this approach the barcode, while the decoding is performed on the
cannot successfully recognize the barcode automatically as grayscale image. Due to thresholding, the map Is(n) may
many threshold values need to be set again, e.g. the local contain more than one blob. Instead of computing the
de-noise threshold. connected components of the thresholded map, we just
select the pixel,No that maximizes Is(n), with the
assumption that the correct blob (the one corresponding to
4 BARCODE LOCALIZATION
the barcode) contains such a pixel. In our research, this
We found out that the simple and fast algorithm presented assumption was always found to be correct. We then
by Orazio Gallo and Roberto Manduchi [4] provides expand a vertical and a horizontal line from No, and form a
excellent results even in challenging situations. This rectangle with sides parallel to the axes of the barcode
localization algorithm assumes that the image of the image and containing the intersections of these lines, with
barcode is captured by the camera-oriented system such the edge of the blob. Note that a quiet zone, a white region
that its vertical axis is approximately parallel to the bars as around the barcode that facilitates localization, borders the
shown in Fig. 3. This image is further converted to a leftmost and rightmost bars of a barcode. The quiet zone
grayscale image as shown in Fig. 4. Thus, with respect to and the large size of the block filter, ensures that the

IJSER
the barcode, one should expect an extended region vertical sides of the rectangle fall outside the area of the
represented by strong horizontal gradients and weak barcode by at least a few pixels as shown in Fig. 6.
vertical gradients. Accordingly, we compute the horizontal
and vertical derivatives, Ix(n) and Iy(n), at each pixel n.
Then we combine them in a nonlinear fashion given as
Ie(n) = |Ix(n)| − |Iy(n)|
It is acceptable to assume that many points within a
barcode should have a large value of Ie(n). We run a block
filter of size 31x 31 over Ie (n), obtaining the smoothed map
Is(n).

Fig. 5.Gaussian filtered Image

Fig. 3.Original Image

Fig. 6.Barcode region Image

Fig. 4.Gray scale Image

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 4, April-2016 85
ISSN 2229-5518
5 IMAGE PRE-PROCESSING 5.4Edge Detection
5.1Cropping An edge is a pixel at which there is a sudden intensity
change. An array of such pixels is created that forms the
When an image is taken for scanning, along with the
edges of the bars of the barcode image. Bar widths are
barcode, other unnecessary information like letters and
calculated by subtracting consecutive elements of the array
words surrounding the barcode also gets included. Thus, it
of edges. These bar widths are given as input to the
is very crucial to remove the noise and get only the barcode
decoding algorithm.
region for proper scanning and decoding. After barcode
localization, only the barcode region is visible and
surrounding information is eliminated by making all other
pixels value to 0. Then, we perform cropping by observing
the intensity of each pixel and extracting the rows of a 6 DECODING ALGORITHM
barcode, i.e. the pixels having the intensity greater than 0. The barcode number is decoded using the array of bar
Fig. 7 depicts the cropped image of barcode form Fig. 6. widths. The first digit is always 8 as it is the first digit of the
country code and is fixed. As from the Table 1, the first
group of 6 digits is encoded as LGLGGL and the last group
of 6 digits is encoded as RRRRR. Each digit is represented
by 4 bars, such that black and white bars are alternate. For
the first 6 digits the first bar is always a white bar and next
black and so on and for last 6 digits it is exactly reverse. A
bar can have four widths from 1 to 4. From the array of bar
Fig.7.Cropped Image widths, the first and last three widths are always one
indicating the starting and ending of the barcode. The

IJSER
remaining widths are grouped 4 at a time as a particular
5.2 Contrast Enhancement digit is represented using 4 bars. In Table 2, 0 represents
In a real-time image of a barcode, the contrast between white, and 1 represents black. For example, if a group of 4
white and black bars is poor due to blurriness.Hence, it is widths (representing a number from the first 6 digits) is
necessary to improve the contrast of the image to 3112, it is considered as 0001011 (3 is the width of a white
distinguish between bars. This is done by making the black bar hence represented by three zeroes). Here, 3 is the width
bars one shade darker on the grayscale compared to white of a white bar, 1 is of black and so on. Thus from the Table
bars as shown in Fig. 8. 1, this first group of 4 widths that is encoded under the L-
code is searched for in Table 2 under the L-code and gets
matched with the code of digit 9. Hence, the first digit is
decoded as 9. Similarly, other digits are decoded using the
table.

Fig. 8.Contrast improved Image 7 DISPLAY OF PRODUCT INFORMATION

Initially, a database of products is created in an excel sheet
using MATLAB R2009b. It is very easy to import and
5.3Converting to ideal barcode image export data from an excel sheet to MATLAB. Barcode data
can be entered into and read from an excel sheet in
The contrast-enhanced image is firstly binarized. But due to
MATLAB conveniently. The database consists of fields such
the real-time acquisition, this image is distorted. To convert
as barcode number, product name, manufacturer's name
into an ideal image, every column of an image is scanned
and product price. Other fields can also be added as per
and checked for a maximum number of pixel intensity of
requirement. (Note that in an excel sheet, the 13 digit
either 0 or 1. If a particular column has more number of
barcode number is not displayed fully in the column but is
pixels with intensity 1 than 0, then the whole column is
visible in the form of an exponential number. To read the
converted to pixels with intensity 1. Fig. 9 shows an ideal
whole number it is necessary to click on that particular
barcode image.
cell.)
After the whole number is decoded, it is compared with the
database and when a match is found the corresponding
information is displayed in another excel sheet. As more
number of barcodes is scanned their information is added
to the same sheet and a bill is created (by adding the price
Fig. 9.Ideal Barcode Image of each product) as shown in Fig. 10.

IJSER © 2016
http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 7, Issue 4, April-2016 86
ISSN 2229-5518
This is an efficient algorithm, which can be further
improved by adding angle correction method and can be
provided to small grocery shops that cannot afford costly
barcode scanners. This saves a lot of time than manual
calculation and also removes the possibility of human error.
We have extensively tested our algorithm on challenging
images and the accuracy is found to be above 60 percent.

REFERENCES
[1] Xianyong Fang, FuLi Wu, Bin Luo, Haifeng Zhao and
PengWang,“Automatic Recognition of Noisy Code-39 Barcode,”
Artificial Reality and Telexistence–Workshops, 2006. (ICAT ’06),
pp.79,82, Nov. 29 2006-Dec. 1 2006,doi: 10.1109/ICAT.2006.45H.
[2] A.S.Bhaskar Raj, Barcodes Technology and implementation,pp.1-15.
[3] Chai D, Hock F, “Locating and Decoding EAN-13 Barcodes from
Images Captured by Digital Cameras,” Information,
Fig. 10.Final Bill Communications and Signal Processing, 2005 Fifth International
Conference,pp.1595, 1599,doi: 10.1109/ICICS.2005.1689328
[4] Thomas, V.M., “A Universal Code for Lifecycle Management of
Products,” Electronics & the Environment, Proceedings of the 2007
IEEE International Symposium, pp.180,183, 7-10 May 2007 doi:
10.1109/ISEE.2007.369390.

IJSER
[5] Gallo, O.,Manduchi, R., “Reading 1D Barcodes with Mobile
8 CONCLUSION Phones Using Deformable Templates,” Pattern Analysis and
We have presented an algorithm for decoding (localization Machine Intelligence, IEEE Transactions, vol.33, no.9, pp.1834,1843,
and reading) barcodes through a webcam, using image- Sept. 2011 doi: 10.1109/TPAMI.2010.229
processing techniques such as filtering, cropping, and [6] Luping Fang; Chao Xie, “1-D Barcode Localization in Complex
contrast enhancement and edge detection. This localization Background,” Computational Intelligence and Software Engineering
technique is fast and accurate even for cluttered images. (CiSE), 2010 International Conference, pp.1,3, 10-12 Dec. 2010 doi:
Barcodes having a flat surface and a width to height ratio 10.1109/CISE.2010.5677072
between 2 to 3 are successfully decoded. We have [7] N. Otsu, “A Threshold Selection Method from Gray-Level
Histograms,” IEEE Trans. Systems, Man, and Cybernetics, vol. 9, no.
implemented this algorithm in MATLAB R2009b and can
1, pp. 62-66, Jan. 1979.
work on any of the later versions of MATLAB. We have
scanned multiple products by executing this algorithm in a
loop, and a final bill of products is created in an excel sheet.

IJSER © 2016
http://www.ijser.org