ArticlePDF Available

Recognition and generation of QR/bar codes in mobile applications

January 2019

January 2019
5(1):36-42

DOI:10.32434/2521-6406-2019-5-1-36-42

Authors:

Larysa Korotka

Ukrainian State University of Chemical Technology

Content uploaded by Larysa Korotka

Content may be subject to copyright.

UDK 519.67:681.332

Nikonenko A.N., Korotka L.I.

RECOGNITION AND GENERATION QR / BARCODE IN MOBILE APPLICATIONS

Ukrainian State University of Chemical Technology, Dnipro, Ukraine

The article is devoted to the analysis of methods of recognition and generation of QR / BAR codes

in mobile applications [1]. The most common algorithms for detecting and coding codes are

considered: Otsu method, Reed-Solomon codes, binary coding algorithm, Orazio Gallo and Roberto

Manduchi approach. In this article, it is proposed to use an algorithm based on these approaches,

which is refined and used for the software implementation of the mobile proposal. The main stages

of the recognition process in the work are: converting a color image into gray shades (using the

pixel hue difference method); gray image binarization (Otsu algorithm); using affine

transformations to preserve parallelism, which includes rotation, stretching, shifting, scaling. The

main stages of the generating process QR codes are: processing and reduction to the standard for

QR codes type; the process of generating a binary string (in the work data of only numerical and

symbolic types were considered); counting the total number of characters in the source string and

converting the decimal number to a binary representation; determining the length of the generated

binary string according to the Denso Wave standard (the encoding of character types is different);

determining the required length of the final binary string; auxiliary tokens are concatenated;

generation of Reed-Solomon error correction codes. Three types of BAR codes are considered: code

128, code 39, and EAN 13. The last type used. A BAR code scanner includes two steps for

automatically scanning all BAR codes: finding the location of the BAR code, BAR code decoding.

To localize the image and the BAR code that is captured by the system, the algorithm of Orazio

Gallo and Roberto Manduchi is used. In order to obtain a smoothed map, a block filter of a certain

size was used, taking into account the size of the input code image. In addition to the code

localization, it is cut off: the intensity of each pixel is taken into account, and the lines of the BAR

code (pixels with an intensity greater than zero) are highlighted. An image with improved contrast is

converted to binary form, and to convert to an ideal image, each image column is scanned and

checked for the maximum number of pixels with an intensity of zero or one. This article focuses on

describing the edge detection process. The decoding algorithm uses an array of code width

bands. All the above approaches and algorithms based on them are implemented as a mobile

application for recognition and generation of codes.

Keywords: recognition, generation, binarization, affine transformations, BAR code.

Formulation of the problem

The use of QR codes (Quick Response code) in everyday life opens up new opportunities: to

establish a connection between hyperspace and real life, using modern technologies for transmitting

digital data streams and methods of indirect information transfer. QR codes are often used for

commercial tracking, travel tickets, storage of personal information, etc.

As you know, BAR codes are read using scanners that use laser beams. Currently, smartphones can

also be used as scanning devices for QR codes. This is a very simple, easy and convenient way to

access specific URLs or websites. The fact that there is no need to manually enter the URL provides

a better conversion, and therefore it is good for marketing purposes and more.

Quite contradictory requirements are put forward for QR codes: the code must store a huge amount

of necessary information, however, at the same time occupy a small area. In addition, if the

information is damaged, then it should be easily restored. In this case, the code must have a high

recognition rate, despite the fact that some of it may be missing. Obviously, it is quite convenient

and practical to scan information thanks to mobile phones on which the recognition program is

installed. Obviously, the development of free mobile applications recognition and code generation is

relevant.

Analysis of recent research and publications

It should be noted that, despite the practical application of codes and the relevance of their use,

there are not so many works on their detection and generation. As you know, the concept of a QR

code first appeared in Japan in 1994: a BAR code standard was developed and presented to track

the different stages of Toyota production at its enterprises. Due to simplicity, the popularity of QR

codes has increased, which made it possible to use these two-dimensional BAR codes in all spheres

of human life [2]. The most common is the algorithm proposed by the Otsu scientist, calculating the

binarization threshold for a halftone image, which is used in the field of computer pattern

recognition and image processing to produce black and white images [3].

One approach to decoding BARcode was proposed by Orazio Gallo and Roberto Manduchi. The

method is based on deformable patterns and uses all the information about the gray level of each

pixel. They presented a new BAR code decoding algorithm (localization and reading) that can work

with blurred, noisy images and low-resolution images [4].

In the general case, when generating noise-proof codes (Reed-Solomon), multiplication and

division of polynomials into each other is used [5]. The algorithm proposed by Luping Fang and

Chao Xie can detect a BAR code area even against a complex background using image analysis

based on areas. Unlike traditional image analysis based on regions, the algorithm is processed using

an integral image. It can work with images that are blurred, contain obliquely arranged BARcode

areas and are shot under different lighting conditions. The fast and reliable way to detect a noisy

BARcode is offered by Xianyong Fang[6-8].

Formulation of the objectives of the article (problem statement)

The paper discusses the methods of recognition and generation of QR / BAR codes that are

available in the public domain and are analyzed. The aim of the work is to develop an algorithm for

generating and detecting codes, on the basis of which a mobile application is created. In this case,

special attention is paid to the performance of the developed algorithm.

Presentation of the main research material

As you know, the use of two-dimensional BAR codes is that the printed code is transmitted to the

object, then read using a mobile device that has a functioning camera and the necessary installed

software that will decode the BAR code itself.

The QR code detection algorithm presented in this work consists of several stages: searching for

outlines in the image; contour approximations; search for suitable quadrangles; FP block ratio

checker (Finder Pattern).

There are not many programs that read and decode QR codes, they are all implemented using a

simple code detection algorithm on the image obtained from the camera, after which a simple

procedure for decorating information takes place. And all this requires a completely clear

positioning of the selected area on the device and a special arrangement of QR codes, the area must

exactly coincide with the faces of the QR codes. That is, the user needs to independently focus on

the necessary QR code and control that the area for scanning the QR code coincides with the QR

code itself, which needs to be decoded.

Let us dwell on the important points of this process. First you need to convert a color image into an

image with gray shades. It uses a method that is based on the difference in the illumination of

pixels. It is a modifiable analogue of the method of averaging the values of the three components of

each pixel [2]. Calculated by the formula: P[i][j]=(R+G+B)/3.

For this work, the selected coefficients of color components: Pnew=0,21*R+0,71*G+0,07*B.

The result will be the calculation of all pixels in the image by the formula, you get a new image,

which will be only in gray.

For binarization of the image, the Otsu method is used, where the image forms a bimodal

histogram. Next, the optimal threshold is computed, which divides the two classes so that their

intragroup dispersion tends to the minimum value.

The method is described in this way [3]:

Let q1 and q2 be the probabilities of each class, which were calculated using the formulas:

   

xPtq

1



   





I

1ti

2xPtq

Dispersion for each of the classes is determined as follows:

   

 

 





 t

1i 1

1tq

*tit

   

 

 





 I

1ti tq

*tit

The average value of each class is calculated:

 





 t

1i tq

iiP

 





 I

1ti tq

iiP

In order to get the result you need to maximize the following equation:

     

 

   

 

2111

btttq1tqt 

As a result, the required threshold was obtained, by which it is possible to divide the histogram into

two classes and to assign the pixels of the smaller class to zeros, and to the pixels of the other class -

units. Thus, the binarized image is obtained and proceed to the next stage.

As is known, affine transformations retain parallelism. The full group of affine transformations

consists of four transformations:

– turn;

– stretching;

– shift;

– scaling.

Transformations of this group can be written in different ways, for example, [4].

The process of coding QR codes is divided into several main stages. We describe this process,

which is implemented in this work.

The first stage is that the line with the input data is processed and reduced to the standard for QR

codes type.

After the string with the initial data is received, the process of generating the binary string begins.

Four technical bits are added to the beginning of the binary string X, which are determined based on

the data type being encoded. The list of possible combinations includes: 0001 – numeric data,

0010 – alphanumeric data, 0100 – binary information, 1000 – special Japanese characters. In order

to simplify the process of displaying the results, only numerical and symbolic data were considered

in the paper.

In the next generation step, the total number of characters in the source string is calculated and the

decimal number is converted to a binary representation. It is necessary to determine the length of

the generated binary string according to the Denso Wave standard.

In order to encode a character data type, the source string must be broken into pairs of characters.

Then you need to know the value from the ASCII table for the first character in each of the pairs,

this value is multiplied by the number 45, then the value is taken according to the same ASCII table

for the second character in each of the pairs. These two values are summed, and each pair of

symbols has its own numerical value. The resulting numbers must be sequentially represented as

11-bit numbers in binary form. If the line contains an odd number of characters, then the value for

this character from ASCII is taken and converted to a 6-bit binary number. All received blocks are

also concatenated with the generated string X on the right side.

After completing the steps described, it is necessary to again refer to the version table of DENSO

WAVE, and determine the required length of the final binary string. If the length of the string X is

less than the required, then from the right side is added from 1 to 4 bits (depending on how many

characters are missing). The generated string X is divided into 8-bit tokens (blocks) and, if the last

block is less than 8 bits, then zeros are appended to it on the right side.

At the final stage of generating a binary string, auxiliary tokens are concatenated with string X. In

case the generated string is smaller than the set (according to the version table) second type, eight

bit blocks are added to the end of the generated string in turn, until the required number of bits:

11101100 and 00010001. After the generated string contains the required number of bits, the binary

string is considered to be finally generated and can be used in the next steps of generating QR

codes.

Next, go to the generation of Reed-Solomon codes for error correction. As noted earlier, in the

general case, when generating Reed-Solomon codes, the multiplication and division of polynomials

are used. For example, if polynomial A is a code word (generated on the basis of the binary string

obtained in the previous step), and B is an irreducible polynomial that is known to both sides

exchanging information, then their multiplication will generate a new polynomial C.

This process means that when the transmitted message is received, the process of decoding it will

be to divide the polynomials into each other. If the division of the polynomial C by the polynomial

B leaves a remainder, then the transmitted message contains an error. The remainder of the division

forms a polynomial syndrome, through which you can find out the damaged bits. In addition, in

some cases, you can correct a certain number of errors in the received message. If the degree of

polynomial B differs from the degree of polynomial A by at least two, then it is possible not only to

determine the presence or absence of an error in the transmitted message, but also to correct some

of them. In this case, the following equation (where k – is the degree of the polynomial; t – is the

maximum value of possible corrections) shows that the redundancy of Reed-Solomon codes is:

k=2*t.

The process of generating Reed-Solomon codes, their encoding and decoding, is described in the

appendix of the QR codes specification [5]. As a result of generating a string using Reed-Solomon

codes for the original message, a set of decimal values will be generated. They need to be converted

to binary format and concatenated with the source line on the right side. The result of all the

previous steps will be a string in binary format containing the original message data plus Reed-

Solomon codes in order to provide the possibility of error correction in the decoded message. Using

Reed-Solomon codes, code words are generated in order to recover errors when reading information

from a QR code. Then they are combined with the original message, which was generated at the

first stage of the algorithm. The result will be one line containing a certain number of bits according

to the specification. At this stage, the QR code has not yet been formed. Next, we analyze how the

two-dimensional representation of the QR code is generated. To begin with, a standard QR code

pattern is generated, in which there are all the necessary technical bits needed to define auxiliary

information. Such a standard template is the basis of any QR code. This template always includes

three positioning markers located in the upper left and right corner, as well as in the lower left

corner of the template. In addition, all the templates contain special Timing Patterns (alternating

white and black pixels alternately), which serve to pre-determine the version of the QR code.

With the known version of the QR code and the level of error correction and the presence of all the

technical primitives in the template, the filling of the template of the QR code with previously

generated information begins. The process of adding bits of the original message to the matrix is as

follows: first, the first pixel (the color is selected according to the value of the bit) is located in the

lower right-most corner of the template. The second pixel is added to the left of the first on the same

line. The next pair of bits is located on top, similar to the first two. This is how the entire template

column is filled. When the filled column has reached the top of the template, the process begins

anew, but from the first free cell located on the right above (the same process, but only from the

bottom up), as well as the first column. Thus, the entire QR code template is filled. The received

QR code is not fully completed at this stage. In order to complete the code generation process, it is

necessary to apply special masks and calculate penalty values from the received QR codes.

According to the specification, it is necessary to generate eight different types of QR codes based

on the newly generated one. Each of these eight codes is defined by its formula, which are

presented in the table 1.

Each of the formulas presented is used to determine whether a pixel located with certain indices

should be replaced with the opposite value.

A BARcode is a visual display of information in the form of stripes and gaps on the surface. Stripes

and spaces can be of different widths and consist of numbers, symbols. Today, different types of

BAR codes are used (code 128, code 39 and EAN 13) [6-8]. In this case, EAN 13 is used.

Table 1

Formulas for generating eight codes

Mask Pattern Condition

0 (i + j) mod 2 = 0

1i mod 2 = 0

2j mod 3 = 0

3 (i + j) mod 3 = 0

4 ((i div 2) + (j div 3)) mod 2 = 0

5 (i j) mod 2 + (i j) mod 3 = 0

6 ((i j) mod 2 + (i j) mod 3) mod 2 = 0

7 ((i j) mod 3 + (i+j) mod 2) mod 2 = 0

Here i – is the row index, j – is the column index.

The BAR code scanner includes several BAR code scanning steps:

– finding the location of the BAR code;

– BAR code decoding.

The numbers are divided into three groups, that is, they form three different sets of numbers. The

first group is encoded using a scheme, where each digit has two possible codings: one with even

parity and the other with odd parity. The first digit is encoded with the choice of pattern. Each digit

in the last group of six digits is encoded using one set of column patterns that are used for the

Universal Product Code (UPC). When the first digit is zero, then the digits in the first group of six

are encoded using patterns that are used for the UPC; because of this, the UPC BAR code is treated

as an EAN-13 BAR code with the first digit shown as zero [6].

In this paper, the algorithm was chosen by Roberto Manduchi and Orazio Gallo, which states that

the BAR code image that is captured by the camera, so that the vertical axis is parallel to the stripes.

Also, this image is transformed into an image in a gray tint [7]. It is necessary to calculate the

horizontal and vertical derivatives Ie(n) and Iy(n) for each pixel n, then they are combined in a non-

linear way, which is calculated by the formula: Ie(n)=|Ix(n)-|Iy(n)|.

If the majority of points in the BAR code have a large Ie(n) value when a block filter is needed,

which is 31x31 over Ie(n), then we will obtain a smoothed map Is(n).

The filter size is selected from the size of the incoming BAR code image and the smallest BAR

code size. As a result, the binary form Is(n) is presented with a threshold, which is calculated using

the simple Otsu method, and the output image. Next, select the pixel No, which tends to the

maximum Is(n). The vertical and horizontal line extends from No and a rectangle is formulated,

which has sides parallel to the axes of the BAR code image.

As you know, when a BAR code is scanned, unnecessary information that surrounds the BAR code

enters the camera. Therefore, when the process of localization of the BAR code passes, the

information surrounding it will be deleted (the pixels are zero). The image is then cropped: the lines

of the BAR code are highlighted (the pixels are greater than zero).

Since the image can be distorted in real time, it needs to be converted to perfect. To do this, you

need to check each column for the maximum number of pixels with an intensity of zero or one. If a

column has more pixels with one, then the column needs to be converted to pixels with a unit

intensity.In order to detect edges, you need to create an array of edges that forms the edges of the

BAR code image strips. The bandwidth is calculated by subtracting the successive elements of the

array of edges (the bandwidth is the input to the decoding).

Decoding occurs in such a way that the BAR code number is decoded using an array of width

bands. The first digit does not change (because the first digit is the country code, it will not change).

The first group, which includes six digits, is encoded as LGLGGL, and the remaining group, which

also consists of six digits, is encoded as RRRRR (the black and white columns alternate between

where white is zero and black is one) [7].

All of the above is implemented as a mobile application, some of the results of its use

are shown in the gure 1.

Fig. 1. Windows of the mobile application "Recognition and generation of QR / BARcodes"

Findings

The article describes an approach to developing a mobile application for recognizing and decoding

QR / BARcodes.

As a result, the following tasks were solved:

– algorithms for the transformation of geometric objects on the plane, coding and decoding of

QR / BARcodes, recognition and image processing were studied;

– algorithms for the selection of QR / BARcode in the image;

– algorithms necessary for converting QR codes to standard form using affine transformations are

presented;

– implemented QR / BARcode decoding algorithm;

– developed a mobile application based on the described algorithm.

REFERENCES

1. Korotka L.I. Mathematical methods for processing multidimensional data / L.I. Korotka,

Nikonenko A. N.// MATERIALS ІІІ-rd International scientific-technical conference COMPUTER

MODELING AND OPTIMIZATION OF COMPLEX SYSTEMS (Dnipro, 1-3 November 2017). –

Dnipro: USUCT, 2017. – P. 116-117. (DOI: http://dx.doi.org/10.32434/CMOCS-2017)

2. ISO/IEC 18004:2015 Information – Automatic identification and data capture techniques – QR

Code BARcode symbology specification. Available at: https://www.iso.org/standard/62021.html

3. ISO/IEC 18004:2000 Information technology – Automatic identification and data capture

techniques – BAR code symbology – QR code. Available at:

https://www.iso.org/standard/30789.html

4. Otsu N. A Threshold Selection Method from Gray-Level Histograms // IEEE Trans. Sys., Man.,

Cyber. Vol. 9, Issue: 1. 1979. – P. 62-66. (DOI: 10.1109/TSMC.1979.4310076)

5. David F. Rogers, J. Alan Adams Mathematical elements for computer graphics // Second Edition.

McGraw Hill Education, 2012. – P. 611.

6. Gill, John Definition of BCH codes // Stanford University. EE387 Notes #7, 2014. – P. 24-86.

Available at: https://web.archive.org/web/20140630172526/http:/web.stanford.edu/class/ee387

/handouts/notes7.

7. Chai D, Hock F, “Locating and Decoding EAN-13 BARcodes from Images Captured by Digital

Cameras,” Information, Communications and Signal Processing, 2005 Fifth International

Conference,pp.1595, 1599,doi: 10.1109/ICICS.2005.1689328

8. Gallo, O.,Manduchi, R., “Reading 1D BARcodes with Mobile Phones Using Deformable

Templates,” Pattern Analysis and Machine Intelligence, IEEE Transactions, vol.33, no.9,

pp.1834,1843, Sept. 2011 doi: 10.1109/TPAMI.2010.229

Стаття присвячена аналізу методів розпізнавання та генерації qr / bar кодів в мобільних

додатках. Розглянуто найбільш поширені алгоритми детектування і кодування кодів: метод

Оцу, коди Ріда-Соломона, алгоритм двійкового кодування, підхід Орацио Галло і Роберто

Мандучі. У роботі пропонується використовувати алгоритм, який базується на зазначених

підходах, та, який доопрацьовано і використано для програмної реалізації мобільного

додатку. Основними етапами процесу розпізнавання в роботі є: конвертація кольорового

зображення у відтінки сірого (використовується метод різниці відтінків пікселів); бінаризація

сірого зображення (алгоритм Оцу); використання афінних перетворень для збереження

паралельності, які включають поворот, розтягнення, зрушення, масштабування. Основними

етапами процесу генерації QR-кодів є: обробка і приведення до стандартного для QR кодів

виду; процес генерації бінарних рядків (в роботі розглядалися дані тільки чисельного і

символьного типів); підрахунок загальної кількості символів в заданому рядку і перетворення

десяткового числа в бінарне представлення; визначення довжини бінарного рядка, що

генерується, відповідно до стандарту компанії Denso Wave (кодування символьних типів

здійснюється інакше); визначення необхідної довжини кінцевого бінарного рядка; допоміжні

токени конкатенуються; генерація кодів Ріда-Соломона для корекції помилок. В роботі

розглянуті три типи штрих-кодів: код 128, код 39 і EAN 13. Використано останній тип.

Сканер штрих-коду включає в себе два кроки для автоматичного сканування всіх штрих-

кодів: знаходження розташування штрих-коду; декодування штрих-коду. Для локалізації

зображення штрих-коду, який захоплюється системою, використаний алгоритм Орацио Галло

і Роберто Мандучі. З метою отримання згладженої карти, використано блоковий фільтр

певного розміру з урахуванням розміру вхідного зображення коду. Крім локалізації коду,

виконується його обрізка: розглядається інтенсивність кожного пікселя і виділяються рядки

штрих-коду (пікселі з інтенсивністю більше нуля). Зображення з поліпшеною контрастністю

перетворюється в двійкову форму для перетворення в ідеальне зображення, кожен стовпець

зображення сканується і перевіряється на максимальну кількість пікселів з інтенсивністю

нуль або одиниця. В роботі окрему увагу приділено опису процесу виявлення границь

зображення. Алгоритм декодування використовує масив смуг ширини коду. Всі вище

викладені підходи та на їх основі алгоритми, реалізовані у вигляді мобільного додатку

розпізнавання та генерації кодів.

Ключові слова: розпізнавання, генерація, бінаризація, аффінне перетворення, штрих-код.

Статья посвящена анализу методов распознавания и генерации qr/bar кодов в мобильных

приложениях. Рассмотрены наиболее распространенные алгоритмы детектирования и

кодирования кодов: метод Оцу, коды Рида-Соломона, алгоритм двоичного кодирования,

подход Орацио Галло и Роберто Мандучи. В работе предлагается использовать алгоритм,

базирующийся на указанных подходах, который доработан и использован для программной

реализации мобильного предложения. Основными этапами процесса распознавания в работе

являются: конвертация цветного изображения в оттенки серого (используется метод разности

оттенков пикселей); бинаризация серого изображения (алгоритм Оцу); использование

аффинных преобразований для сохранения параллельности, которые включают поворот,

растяжение, сдвиг, масштабирование. Основными этапами процесса генерации QR-кодов

являются: обработка и приведение к стандартному для QR кодов виду; процесс генерации

бинарной строки (в работе рассматривались данные только численного и символьного

типов); подсчет общего количества символов в исходной строке и преобразование

десятичного числа в бинарное представление; определение длины генерируемой бинарной

строки согласно стандарту компании Denso Wave (кодирование символьных типов

осуществляется иначе); определение необходимой длины конечной бинарной строки;

вспомогательные токены конкатенируются; генерация кодов Рида-Соломона для коррекции

ошибок. В работе рассмотрены три типа штрих-кодов: код 128, код 39 и EAN 13.

Использован последний тип. Сканер штрих-кода включает в себя два шага для

автоматического сканирования всех штрих-кодов: нахождение местоположение штрих-кода;

декодирование штрих-кода. Для локализации изображения штрих-кода, который

захватывается системой, использован алгоритм Орацио Галло и Роберто Мандучи. С целью

получения сглаженной карты, использован блочный фильтр определенного размера с учетом

размера входного изображения кода. Помимо локализации кода, выполняется его обрезка:

рассматривается интенсивность каждого пикселя и выделяются строки штрих-кода (пиксели

с интенсивностью больше нуля). Изображение с улучшенной контрастностью преобразуется

в двоичную форму и для преобразования в идеальное изображение, каждый столбец

изображения сканируется и проверяется на максимальное количество пикселей с

интенсивностью ноль или единица. В работе отдельное внимание уделено описанию

процесса обнаружения краев. Алгоритм декодирования использует массив полос ширины

кода. Все выше изложенные подходы и на их основе алгоритмы, реализованы в виде

мобильного приложения распознавания и генерации кодов.

Ключевые слова: распознавание, генерация, бинаризация, аффинные преобразования,

штрих-код.

Информационный блок

Recording and generalization QR/BAR code in mobile applications