READING BARCODES USING DIGITAL CAMERAS ... - KULIS

Proceedings of 5th International Symposium on
Intelligent Manufacturing Systems, May 29-31, 2006: 835-843
©Sakarya University, Department of Industrial Engineering
Authors:
READING BARCODES USING DIGITAL CAMERAS THROUGH IMAGE
PROCESSING
Author : Emre BAŞARAN
E-Mail : emrebasaran45@gmail.com
Tel : 0536 257 13 88
Author : Özgür ULUÇAY
E-Mail : ozgur.ulucay@oberonsystems.com
Tel : 0532 526 02 50
Author : Assoc. Prof. Dr. Sarp ERTÜRK
E-Mail : sertur@kou.edu.tr
Tel : 0262 3351148 / 2204
About the Authors:
Emre Başaran was born on 13th May 1983 in Balıkesir, Turkey. He is a last year student in
the Electronic and Telecom Engineering department of Kocaeli University. Since the last year
he has been working on a project based on barcode reading with image processing
techniques. He enjoys computer programming with C, C# and is interested in new
technologies like mobile communications. He wants to continue his studies in signal
processing over mobile communications.
Özgür Ulucay was born in 5 May 1980 in Eskişehir, Turkey. He graduated from the Electronic
and Tel. Engineering department of Kocaeli University. He gained the Master degree in 2004
and continues towards the PhD. degree in the same department. He enjoys computer
programming and studies 3-D modeling and signal processing. He has some papers in the
computer graphics field. He continues to study in signal processing over internet and mobile
programming technologies.
Sarp Ertürk received his B.Sc. in Electrical and Electronics Engineering from Middle East
Technical University, Ankara in 1995. He received his M.Sc. in Telecommunication and
Information Systems and Ph.D. in Electronic Systems Engineering in 1996 and 1999
respectively from the University of Essex, U.K. From 1999 to 2001 he carried out his
compulsory service at the Army Academy, Ankara as Lecturer in the Electrical Engineering
department. Since 2001 he has been with the University of Kocaeli, Turkey, where he is
currently appointed as Associate Professor. He has established and is still directing KULIS
(Kocaeli University Laboratory of Image and Signal processing). He has been offered a
Distinguished Visiting Professor position by Chung-Ang University, South Korea, between
March-September 2006. His research interests are in the area of digital signal and image
processing.

Abstract:
All people go to the supermarket to buy something that needs lots of time. Each product has
a label made up of lines with different thicknesses. We go to cashbox when buying these
products. Cashiers pass the products through a hand-scanner to see the corresponding price.
In this paper, a method is presented for barcode reading with camera which is based on
image processing. Normally, the barcode reading process is performed by laser readers, and
laser readers are necessary tools for barcode readings. In this paper we present an
alternative method for barcode reading directly from camera images. The advantage of the
proposed method is that barcodes can be read directly from the line, having product barcodes
look upwards facing a camera located above the line, so that it will not be necessary to pick
up all individual products to pass them through a laser scanner manually. Thus the proposed
approach will provide speed and simplicity.
Introduction:
Generally, barcodes are symbols shaped in the form of rectangles which consist of thin or
thick parallel lines parallel to each other. Barcodes provide means for automatic rapid data
input into the computer. Since the last decade, barcodes are being used in many areas such
as market products and electronic devices. The lines on barcodes contain the reference
number of the product. This information should be recorded in computers to store each
product separately for counting company sales and purchase quantities. When reading
barcodes on products using some laser scanning device, a signal is generated by the system
and is processed in the computer by some software. Then this information is used to
determine which product is selected. This process provides rapid and reliable sales
opportunities to companies for selling their products.
There are very different types of barcodes: EAN, EAN-13, EAN-8, Code 39, Code 93, Code
128, and UPC are well known. UPC and EAN types are most commonly used cases. The
UPC numbering system is used in Canada and America, while the EAN 13 numbering system
is used in Europe and Turkey [3].
Figure 1: A sample barcode in EAN 13 numbering system
836

Figure 1 shows a sample barcode in the EAN 13 numbering system structure. The black lines
on the top of the barcode represent the logic true or ‘1’, and the blanks represent the logic
false or ‘0’. The width of the thinnest black line is the reference width. The reference width is
presented by a single bit and the largest line width of the barcodes can be four times the
reference width at the maximum. In the same way, the width of the thinnest white line has a
reference value, and the thickest white line can have a width of four times the reference width
at maximum. There are starting and finishing codes equal to 101 values at the beginning and
end of the barcode which show the reference widths. There is a longer barcode giving the
01010 value in the center of the barcode and this code lines have reference widths, too.
Barcodes are formed in the form of a bit string according to line thickness to analyze a
barcode. There are 12 numbers on the barcode and each number is defined by 7 bits.
Barcodes also have 3 start, 3 stop and 5 bits in the middle for referencing. As a result the
barcodes are defined by 95 bits in total.
Barcode values are not obtained directly from these 95 bits long codes. These values will be
obtained using barcode numbering schemes. In this paper, the EAN 13 numbering scheme is
used to determine the barcode reading process. The methodology to achieve the value of the
barcodes using this scheme includes the following structure:
1. Find reference width using start (3); stop (3) and center (5) guard bits.
2. Remove the reference bits (3 start, 3 stop, and 5 center bits) from the barcode code bit string.
As a result obtain the 84 bits long barcode information code bit string.
3. Divide the barcode bits into two areas. The first area consists of the first 42 bits, and the
second area consists of the last 42 bits.
For the first area:
4. Find first the decimal value parity of the barcode. This parity value will be used to determine the
order of barcode code readings. The first 42 bits will be read using the encoding table shown in
Figure 2b according to the order obtained from the odd-even table shown in Figure 2a.
5. Select seven (7) bit to encode/decode a digit:
a. Find the resulting value by searching the respective odd/even column of the left-hand
encoding table shown in Figure 2b.
b. Go to step 5 if remaining bit count is bigger than zero.
For the second area:
6. Select seven (7) bits to encode/decode a digit:
a. Find the resulting value by searching the right-hand encoding table shown in figure 2b.
b. Go to step 6 if remaining bit count is bigger than zero.
7. Construct the barcode code using the obtained values.
Algorithm 1: Barcode encoding algorithm for EAN 13 numbering system.
837

a) Odd/Even Parity Table b) Encoding Table
Figure 2: Encoding tables for the EAN 13 numbering system.
As the first decimal number (for instance the first 9 in the barcode shown in Figure 1) shows
the corresponding row of the odd/even parity table and is not encoded in the barcode, it
cannot be extracted by identifying the thickness of vertical lines. In laser scanner systems,
this number is obtained using parity calculation. For simplification, the parity calculation step
is omitted in the proposed system, and both sides of the left-hand encoding approach shown
in Figure 2b is entered into the database and the correct value is determined by comparing
against both (odd and even parity) cases.
Approach:
Our approach contains an edge detection algorithm to obtain barcode borders from images
acquired using a camera, and some threshold mechanism to process images as binary
patterns.
At first, an image that contains the barcode information is acquired using a camera. The
color image contains in fact the full usable information. For faster and simple processing, the
image is converted to grayscale format. Because the acquired image will typically contain an
area larger than the barcode, it is initially required to crop the barcode area as the rest of the
image is unnecessary.
Because only the barcode region is required, the borders of the barcode line coordinates
must be determined. An edge detection algorithm can be used for determining the borders.
Edge detection algorithms can use different kernels. The Canny edge detector provides a
reasonable and successful approach for edge detection [1]. The Canny detector uses two
kernels, one for the vertical edges and one for the horizontal edges. The utilized kernels are
shown in Figure 3. The Canny algorithm provides an optimal edge detector based on a set of
838

criteria which includes finding the most edges by minimizing the error rate, marking edges as
closely as possible to the actual edges to maximize localization, and marking edges only
once when a single edge exists for minimal response [2].
Figure 3: Canny edge detection kernels. The left kernel is used for the detection of vertical
edges and the right kernel is used for the detection of the horizontal edges [2].
In this paper Canny edge detection kernels are used to determine the barcode area. As a
barcode is made up of vertical lines, it has been found that utilization of the vertical edge
detection kernel is sufficient. Hence the computational load is reduced by using only the
vertical edge detection kernel instead of both kernels. Hence, a single pass with the vertical
edge detection kernel is used to determine the barcode area. After applying the edge
detection kernel, a threshold algorithm is applied on the resulting image. At this stage, the
image indicates all vertical lines including barcode lines.
The algorithm that is used in this work is described below in algorithm 2. Note that
For all pixels:
1. Select the pixel to be processed and determine the kernel area.
2. Apply the vertical edge kernel shown in figure 3 (the left picture) to the selected kernel area.
3. If the result is positive, the new pixel value will be the resulting value.
4. If the result is negative, go to step 1.
5. The average value of the output is calculated.
6. Threshold the output using this average value.
This algorithm gives all vertical lines in an image. This process is shown for example barcode
images in Figures 4-6 below.
Figure 4: An image with purple background color.
839

Figure 5: An image with white background color.
Figure 6: An image with red background color.
The obtained white lines on the resulted image after Canny edge detection and
thresholding will then be separated into several parts. This partitioning is accomplished by
finding neighbor pixels in the image. All neighbor white pixels must stay together, and all
pixels that are not-neighbors must be separated. All of the resulted neighbor pixels will be
stored in the form of neighbor tables as an array. This operation is summarized in algorithm 3
given below.
For all pixels:
1. Select the pixel to be processed.
2. If the pixel is black go to 1. Else go to the next step.
3. If any neighbor table exists, go to step 4. Else construct a new neighbor table entry.
4. For all neighbor tables
a) Search all pixel coordinates in the table to detect any neighbor relationship with
a selected pixel.
b) If any relationship is found, put this pixel in this table and return to step 1.
c) If any relationship is not found, construct a new table and put this pixel in it. Go
to step 1.
Algorithm 3: Constructing neighbor tables.
These tables include all the barcode line places and unnecessary noise. For eliminating
unnecessary noise from the barcode information we have to determine the differences
between them. To do that, at first we must determine the table entries that only includes
barcode line places. The size of tables which include barcode lines must be equal or very
close to each other; because all barcode line heights are equal in an image.
840

There are 30 black stripes and 29 white stripes in every barcode. Our edge detection
algorithm just detects the passing from black to white, so there will be 30 tables which have
same or very close sizes.
In this approach, the 30 tables which contain the barcode line pixels must be consecutive. At
first we must determine the barcode upper-left pixel coordinate which is mainly any one of the
available table entries. Then we have to search all tables to determine the barcode area. To
do that, we use a standard deviation value to find all 30 barcode lines which follow each
other, and have the same or close size. This approach is described in algorithm 4 below.
For all tables:
1. Determine the standard deviation of region areas
2. Construct one counter that counts neighbor tables which have suitable standard deviation
and set counter value to zero.
3. Select one table for comparison of table sizes if there is no selected table.
a) Get the table size.
4. Select next table for comparison.
a) Get the second table size.
5. Compare the values using computed standard deviation.
a) If division of table size value difference is below the standard deviation.
Increase counter value and assign the first table as the one with this counter value.
I. If the counter value is equal to 30 Go to step 6.
II. Else go to step 4.
b) Else go to step 2.
6. First and second selected tables together indicate the barcode start and stop positions,
respectively. The area covering the pixel values of the first and second table is the barcode field.
Algorithm 4: Determining the Barcode Area
The approach given in Algorithm 4 is used to obtain the barcode fields in the sample images
shown in Figure 4-6. In fact, these fields indicate the coordinates of the barcode fields. These
coordinates will be used to obtain the real barcode lines. The lines that are included in the
tables are not the actual barcode lines yet, they only indicate the barcode line positions. The
barcode images only contain barcode lines and their threshold equivalents are obtained using
these coordinates as shown in Figure (7-9) below.
Figure 7: The barcode field of the full image shown in Figure 4.
841

Figure 8: The barcode field of the full image shown in Figure 5.
Figure 9: The barcode field of the full image in Figure 6.
After barcode fields are cut out from the image, a threshold is applied on these fields using
the mean values of these fields. Then, the resulting image only contains the barcode lines
which indicate the barcode codes. The barcode codes are obtained using these images using
an algorithm that stores white and black lines together in arrays by determining positions of
these lines described above. Then, every barcode code is obtained using these array sizes.
Note that the reference sizes have to be considered to make codes meaningful. This
operation is given in Algorithm 5 below.
1. Construct a variable ‘refsize’ to indicate a reference size, and a variable ‘result’ for the result
code.
2. Select first array to be processed.
3. This array should contain black pixels. The refsize variable is set to this array size.
For all lines:
4. Select next array if any unselected array exists else go to step 8.
5. Divide this array size to refsize. The calculated floating value is stored as a variable
‘codeValbyRef’. Round this value to the closest integer.
6. If this array includes black pixels write logic ‘1’ for the corresponding amount of the
codeValByRef value to the result value. Go to step 4.
7. If this array includes white pixels write logic ‘0’ for the corresponding amount of the
codeValByRef value to the result value. Go to step 4.
8. The result value must be the 95 bits long barcode code.
Algorithm 5: Determining the Barcode Code
842

After obtaining the 95 bits long barcode code, it must be converted to meaningful decimal
numbers. This can be accomplished using Algorithm 1, as described earlier. The resulting
decimal numbers indicates the barcode code and is meaningfully constructed as shown in
Figure 10 using algorithm 5 and 1 for sample images shown in Figures (7-9)
Figure 10: The barcode results obtained for the sample images
A total of 15 barcode images with differently sized barcodes have been processed with the
presented method. A 100 % correct detection performance has been obtained for the
method. In the future it is planned to evaluate the methods on a larger database, with images
taken under different lighting conditions, with various orientations and multiple barcodes
within an image.
Conclusions:
In this paper; a method for barcode reading with a camera based on image processing has
been presented. First of all; edges are detected in the image using an edge detection method
and then white areas are assigned into arrays. Then we subtract the barcode area from the
original image using these array sizes. A threshold is applied using images mean values of
the barcode field after this process. Then we get the barcode number from black and white
areas obtained after threshold process. The barcode reading process has been carried out
fast, effectively and successfully with the presented method.
References:
[1] Canny, J., “A Computational Approach to Edge Detection”, IEEE Trans. Pattern
Analysis and Machine Intelligence, 8:679-714, November 1986
[2] Hong Shan Neoh, Asher Hazanchuk, “Adaptive Edge Detection for Real-Time Video
Processing using FPGAs”, GSPx 2004 conference paper:
http://www.altera.com/technology/dsp/conf-papers/dsp-conf-papers.html
[3] Barcode toolkits from Softek Software: http://www.bardecode.com/
[4] Vault Information Services LLC: http://www.barcodeisland.com/ean13.phtml
843