Content uploaded by Kishor Datta Gupta
Author content
All content in this area was uploaded by Kishor Datta Gupta on Mar 16, 2018
Content may be subject to copyright.
59
Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
Kishor Datta Gupta
Department of Computer Science
Lamar University, Beaumont, USA
400 S M L King Jr Pkwy, Beaumont, TX 77705, United States of America
Tel. +1 409-880-7011
kgupta@lamar.edu
Md Manjurul Ahsan
Department of Industrial Engineering
Lamar University, Beaumont, USA
400 S M L King Jr Pkwy, Beaumont, TX 77705, United States of America
Tel. +1 409-880-7011
mahsan2@lamar.edu
Dr Stefan Andrei
Department of Computer Science
Lamar University, Beaumont, USA
400 S M L King Jr Pkwy, Beaumont, TX 77705, United States of America
Tel. +1 409-880-7011
s.andrei@lamar.edu
Abstract
Quick Response Code has been widely used in the automatic identification fields (Liu, Ju, &
Mingjun, 2008). The present work illustrates an image processing system able to discover, split and
decodes the most common 2D symbol used in bar code applications. The different symbol is
processed by manipulating their similarities, to achieve an integrated computational structure
(Ouaviani, Pavan, Bottazzi, Brunelli, Caselli, & Guerrero, 1999). There is not enough novel
approach which could be effective for data transferring to alter various sizes, a little noisy or
damaged and various lighting conditions of bar code image. We proposed a faster QR code which
has more storage and can scan faster. The new QR code generation takes twice the time than normal
QR code but it can also store double QR code data. It's scanning is as faster as other QR code
scanning technique. we deducted the level of color percentage despite positions of color bits in that
module, which increased its scanning speed and reduced noise error rate to less than 10 percent.
Keywords: Quick response code, recognition, Scanning technique.
1.Introduction
In 1994, A Japanese Hardware Company first introduced “Quick Response Code”, which is
known as QR codes. Back then, it was designed to allow high-speed component (Furht, Borko
,2011). Currently, it is used not only for getting information of commercial products but also used
for smartphone tagging. Beginning of computer age, it was necessary to have machine readable
medium or automated data medium, by automated data medium we meant a form of something
which could store data that only a computer can read. There are many forms available but here we
will discuss only one of the optical medium. It is printed data matrix generated image on any
substance. The first example of it is Barcode. The Barcode was influenced by Morse code, its
primary use was for tagging each product (Figure 1).
BRAIN – Broad Research in Artificial Intelligence and Neuroscience, Volume 9, Issue1 (February, 2018), ISSN 2067-8957
60
Figure 1. A sample Linear Barcode
In starting all barcodes was 1D or known as linear, but still, there are many standards and
many types of barcode. To increase data capacity and reduce the error rate, 2D barcode was
introduced. The most popular one of them is QR code. But QR code also has the limitation of data
capacity so researcher introduced High Capacity Color Barcode. However, the introduction of color
in QR codes created new challenging problems. As consequence, the color barcode concept did not
get too much popularity.
Figure 2. 1D barcode examples a) 2d postal barcode b) Colored railroad barcode c) code 49 barcode d) The
Universal Product Code
The 1D or linear bar codes are mainly based on the width of each line by a predefined
standard width. For the width based barcode an example is shown in Figure 2 (d).
Here, the big limitation is for more data barcode width size will get bigger. As we can see
barcode to bitwise waveform transfer in Figure 3.
Figure 3. Barcode width read
For this reason, the linear code is used to only store small code words data, such as product
number or country code. To store more data, 2D barcode was introduced, for example Figure 4
shows a 2D barcode. Advantages of 2D barcodes over linear barcode are not only storage of more
information, but also reduced the error rate, the data encryption, the much less space, the possibility
K. D. Gupta, M. Ahsan, S. Andrei - Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
61
to scan in all directions and printing in curve area.5 This makes 2D barcodes more usable in all
fields. When Microsoft introduced High Capacity Color code (Figure 4(d)) it comes with more
storage and more features. However, it also faces some image processing issues. As other standard
2d codes are based on black and white, this does not depend that much on scanner quality or printed
material quality. As a result, Microsoft discontinued its research on color codes. Several attempts to
solve this problem taken by many other researchers notably by Homayoun Bagherinia and Robert
Manduchi in 2011 (Bagherinia Homayoun and Roberto Manduchi, 2011).
Figure 4. 2D barcode examples a) MaxiCode b) CrontoSign c)ShotCode d)HCCB e)QRcode f)HCCBQ
2. Standard QR code:
To the human eye, all barcodes will look very similar. But if we look it more closely we will
see some differences, although without most trained eye we cannot recover any meaningful data.
This is because all data are generally encrypted. In linear barcodes, normally binary data or symbol
for each letter of English alphabet and number are used. For 2D code, we normally directly use
binary codes of respective data. Information in QRCodes can split up in 6 parts as shown in Figure
5. As like as: Quiet zone, Finder patterns, Alignment pattern, Timing pattern, Version information,
Data cells. (Hartley, Richard, and Andrew, Zisserman,2003)
5 "How QR codes (and other 2D barcodes) work." Explain that Stuff. July 02, 2016. Accessed June 17, 2017.
http://www.explainthatstuff.com/how-data-matrix-codes-work.html
BRAIN – Broad Research in Artificial Intelligence and Neuroscience, Volume 9, Issue1 (February, 2018), ISSN 2067-8957
62
Figure 5. QRCode Data blocks information
3. Our Proposed QR Method
Based on our work on QR code scanning, we think there could be new QR code possible
which able to store more data. Our idea is use half the area of black bit instead of full area. It has
faster decoding and encoding speed.
3.1. Encoding Method
First, we will convert data to normal QR code, then the black bits we will make them one
third instead of full, and place in the middle, then we prepare another QR code same way, rotate
that in 90 degrees and place it on first QR code. And we should add a finder-bit so while decoding
we get the info of decoding bit size, which is essential to start decoding method (Yang, Zhibo,
Huanle Xu, Jianyuan Deng, Chen Change Loy, and Wing Cheong Lau, 2017).
Figure 6. Encoding QR code step1
In our Experimental setup, we placed a black bit in top-left most corner. In Figure 6 and
Figure 7 these encoding steps are explained.
K. D. Gupta, M. Ahsan, S. Andrei - Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
63
Figure 7. Encoding QR code Step 2
3.2. Procedural steps for encoding
Steps of encoding are as follows:
Step 1. Convert the data in Binarystring consisting only 1 and 0.
Step 2. Split the Binarystring in two part, part1 and part2
Step 3. Add ending bit 1 in the end of both part1 and part2
Step 4. Divide imagesize with the SQRT of Part2 data length. This will be the bitsize
Step 5. Place a Rectangle in BitSize at the top-left most corner in image.
Step 6. Take another two samesize bitmap.
Step 7. In bitmap1, Take each character from part1 and started to draw line if value which width is
one third of the bitsize but length is equal to bit size. After reach at image end increase X position by
bit-size and start again.
Step 8. In bitmap2, Take each character from part1 and started to draw line if value which width is
one third of the bitsize but length is equal to bit size. After reach at image end increase X position by
bit-size and start again.
Step 9. Merge bitmap1 and bitmap2 and get the image
The image we get from step 9 is our desired new QR code. Based on image size, our bit
limit increase and decrease. If we use 1024 as image size, theocratically we will be able to keep
262144 bit data in our code, but Technically it betters to keep one fifth of that to avoid decoding
issues.
3.3. Pseudocode for encoding
The pseudocode of the encoding process is explained below and details source code are
available in appendix A.
pseudocode:
var result = StringToBinary(InputText);
var part1 = first half of result + "1";
var part2 = second half of result + "1";
var sqrt =sqrt(part2 data size);
var bitsize = imagesize/sqrt;
var x = sizeofbit / 3;
var y = 0;
foreach (char c in part1) {
if(c==1) var bitmap1 = draw line in x and y position with width of bitsize/3 and
lenght of bitsize;
if(y>=imagesize) {y = 0;
x = x + sizeofbit;} }
y = sizeofbit / 3;
x =0;
foreach (char c in part2) {
if(c==1) var bitmap2 = draw line in x and y position with width of bitsize/3 and
lenght of bitsize;
if(y>=imagesize) {x = 0;
y= y+ sizeofbit;}}
var QRcodeimage = new image (imagesize + bitsize,imagesize + bitsize);
QRcodeimage draw rectengale at 0,0 point with bitsize as height and width;
QRcodeimage =draw bitmap1 and bitmap2 from bitsize,bitsize location;
QRcodeimage save at filelocation;
BRAIN – Broad Research in Artificial Intelligence and Neuroscience, Volume 9, Issue1 (February, 2018), ISSN 2067-8957
64
3.4. Implementation of Encoding
We implemented in visual studio. We use default c# image library for decoding purpose and
after that, we save the image in jpeg format. In Figure 8, we paste a text from CNN news at our
inputbox and in Figure 9 we converted it to binary, third image is our desired QR code.
Figure 8. Encoding in visual studio for datainput.
Figure 9. Generate QR code
We can see in figure 3, in topleft corner we placed bit size. This we will use as finder pattern
in decoding steps. After that, we can save the image as QR code in the file directory. We used 1024
as image size. Add a rectangle for of size-bit add extra size, based on length of data bit. Time to
encode normally done below 100ms for normally 300 to 500 words, largely based on data content,
as for 0 we don’t use draw functions.
K. D. Gupta, M. Ahsan, S. Andrei - Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
65
4. Experimental Results of Encoding
We ran several tests on our application, and save the resulting image in our file directory.
Also, we measure the processing time for encoding method. In Table 1, we showd some of the
examples alongside results and processing time in milliseconds.
Table 1. Encoding Experimental result
Input Text Output QR in 1024size Process time in Milliseconds
“Hello World”
75
“Lamar University Kishor Datta Gupta
L20421951”
73
Abstract of this paper
103
Page 1 of this paper
115
Article
from“http://artssciences.lamar.edu/computer-
science/news/lamar-university-virtual-
map.html”
555
5. Analysis of Encoding
From the Table, we can see if data size increases our encoding speed also decrease. For more
data, we will be needed to increase our image size. The Zbar and Zxing QRcode reader software
need more time than our encoding method, mostly because they need to draw the rectangle and fill
that while are drawing the only line which gives us the better performance as we are drawing on
sorted image matrix.
Our time complexity is T(N) = T(N/2) +T(N/2) => O(n) = N
One of the issues in the Encoding process is rounding up the floating bit size to integer
point, As we flooring the results, we have to waste some space. If data is big, then this waste of
space gets more visible, In Figure 10 there is an example shown.
BRAIN – Broad Research in Artificial Intelligence and Neuroscience, Volume 9, Issue1 (February, 2018), ISSN 2067-8957
66
Figure 10. Waste of space in right bottom corner
In Figure 10 we can see due to floor the float value to nearest integer value creates a blank
space in right bottom corner. For decoding purpose we mark last bit as black one, so after the last
data, we have to keep all white. In future, we could use this space to store error correction bits.
Another possibility is brought that in the center and use for curvature detection.
5.1. Decoding Method
We can apply bar code decoding system here that’s why it will be much faster, and we will
ignore black length less than the module size. This way it would be faster than other QR code, we
have to scan horizontal and vertical and ignore lest than module size black color block as 1. As seen
in Figure 11 and Figure 12.
Figure 11. Decoding vertical scanning
Figure 12. Decoding Horizontal scanning
K. D. Gupta, M. Ahsan, S. Andrei - Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
67
5.2. Procedural steps for decoding
In decodings, steps can be divided into three parts. One gets the bit size from processed
image, second is scan and get the data bits, third and last is decode the data in text, in below
pseudocode step 1 to 4 is for getting the bit size and 5 to 8 is for scanning purpose.
Steps of decoding are
Step 1. Get the input image .
Step 2. Convert the image in greyscale.
Step 3. Convert input image in binary image. Using threshold value 128.
Step 4. Started from top left corner to width or height, check every pixel value. If get 255 , that pixel position
represent the bitsize
Step 5. Start to go pixel by pixel horizontally and check blackbit are equal or atleast 3/4th of bitsize or not. If
yes we will add that as 1 bit . if not that will add as 0 bit.
Step 6. If it is the the most right bottom pixel we go step 7. Else after reach the end of image height, will
increase position of X by bitsize and Go to Step 5.
Step 7. Start to go pixel by pixel vertically and check blackbit are equal or atleast 3/4th of bitsize or not. If yes
we will add that as 1 bit . if not that will add as 0 bit.
Step 8. If it is the the most right bottom pixel we go step 9. Else after reach the end of image width, will
increase position of X by bitsize and Go to Step 7.
Step 9. Convert the data from binary to ASCII.
After step 9, we will get our resulted text in ASCII format. We have to remember before
converting to ASCII we have to remove every bit after last 1. As that’s the ending flag bit.
5.3. Pseudocode for decoding
The pseudocode of the Decoding process is explained below and details source code is
available in Appendix B.
Pseducode for Sizebit
var image = QRcodeimage
var greyscaleimage = Convert QRcodeimage to Greyscale
Vat Bbinaryimage =Convert Greyscaleimage to binary using threshold 128;
for (int j = 1; j < img.Height; j++){
var pixel = GetPixelat 1,j point;
if (pixel == 255){bitsize = j; break;}
}
After getting the bit size we can start scanning the image file and decoding to binary.
Pseducode for Decoding
for (int i = bitsize +bitsize/3; i < img.Width; i = i + bitsize; ) {int
ptb,ptw,bitsizecount = 0;
for (int j = bitsize ; j < img.Height; j++) { bitsizecount++
int pixel = Bbinaryimage.GetPixelat 1,j point;
if (pixel == 255){ ptb++;ptw = 0; }
if (pixel == 0) {ptb=0;ptw++; }
if (bitsize == bitsizecount) {
if (ptb >= ptw) horijontal = horijontal + 0;
else horijontal = horijontal + 1;
ptb,ptw,bitsizecount = 0;}} }
for (int j = bitsize + bitsize / 3; j < img.Height; j = j + bitsize;) { int
ptb,ptw,bitsizecount = 0;
for (int i = bitsize ; i < img.Height; i++) { bitsizecount++
int pixel = Bbinaryimage.GetPixelat 1,j point;
if (pixel == 255){ ptb++;ptw = 0; }
if (pixel == 0) {ptb=0;ptw++; }
if (bitsize == bitsizecount) {
if (ptb > ptw) vertical = vertical + 0; else vertical = vertical
+ 1;
ptb,ptw,bitsizecount = 0; }} }
var binaryresult = horijontal+vertical;
Var result = binrayresult to ascii convertion
BRAIN – Broad Research in Artificial Intelligence and Neuroscience, Volume 9, Issue1 (February, 2018), ISSN 2067-8957
68
5.4. Implementation of Decoding
We use default Aforge image library for decoding purpose and after that, we use normal file
reader to scan the image. In Figure 13, we load a QR code from our file directory and in Figure 14
we decoded it to binary and convert that binary in the text, Results are shown in two textboxes.
Figure 13. QR code load in application
Figure 14. QR code decoded and result shown.
After image convert to binary, we start for determining the size bit. In Figure 15 at top left
corner, sizebit is present. Using that sizebit we started to scan the image.
K. D. Gupta, M. Ahsan, S. Andrei - Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
69
Figure 15. Decoding with grid line
For better understanding, We draw redline while scanning in Figure 15 when we get the
constant black line which length is almost equal to bit size, we will consider that as 1 otherwise,
that’s a 0. We also have to remember some digit does convert to binary ok, and it still does not
support other than ASCII code.
6. Experimental Results of Decoding
We ran several decoding tests from the QR code which we encoded and saved in our file
directory. We calculated the processing time for encoding method. In below Table, we showed some
of the examples alongside results and processing time in milliseconds.
Table 16. Decoding Experimental result
Input Image Output Processing time ms Data length
“Hello world” 32 96
“Lamar University” 26 136
Thesis name 12 536
Abstract of this thesis 210 8800
BRAIN – Broad Research in Artificial Intelligence and Neuroscience, Volume 9, Issue1 (February, 2018), ISSN 2067-8957
70
Acknowledgment of
this thesis
165 6424
7.Analysis
The decoding is much faster than encoding, and also compare to other QR code decoding
systems. Its speed is faster as Linear bar code. Its time complexity depends on bitsize. If N = image
size/bitsize . then time complexity is 2T(N) => O(N). Also as lines are sequential we do not have to
worry about cross color modulation error. Also as we are using one-third area of the bit module, it
has more error prone than other QR code scanning, that’s why in this case our zone based error
detection system will be helpful as it will work none the less the printer or printing materials faults.
7.1. Advantages of new QR code
There are several new advantages of our proposed QR code over currently most used QR
codes, These are
Our QR code can store almost double data than Universal QR code.
Our QR code encoding and decoding time are faster than Universal QR code.
Our QR code can detect curvature distortion issue better.
Cross module color issue can be ignore .
As we are reduced the size of blackbits one third , it giving us the option to add more data
than normal QR code. Also while encoding instead of draw a black rectengualr bit , we draw a
single line. Also in a 2d matrix which is sorted. It makes the encoding speed faster . In decoding
time, we use to scan in single line. So its take less time than othe QR code which read a full
rectengular bit and determine its color to measure its black bit or white bit. While we are just chking
for black line. This procedure also help us to avoid curve distortion and cross module distortion
issue.
7.2. Limitations of New QR code
Our new QR code has some big limitations , these are
For large data if image size small it will not work.
If image size can not be divisible by bitsize , for large data size we will be lost line after read
some data.
This new QR code issue is more vulnarbel for light glare and illumination issue.
For solving, these issues we need to increase image size respect to data size, and we can
always add garbage data to keep bit size a divisor for image size. In future, we can try to solve these
issues.
8. Conclusion
In our new QR code method, we can also add the third and 4th layer of QR code by rotating
45 and 135 degrees respectively, that’s why it can store 4X data than normal QR code. And if we
apply HiQ color bits instead of binary bits, that could make this QR code a huge storage of database
to use securely. Also, it needs to make applicable for camera work, especially for mobile camera.
Another room of improvement is, adding curvature detection bits. If we could do this than it will be
possible to make it user-friendly in consumer level.
References
Bagherinia, Homayoun, and Roberto Manduchi. "A theory of color barcodes". In Computer Vision
Workshops (ICCV Workshops), 2011 IEEE International Conference on, pp. 806-813.
Furht, Borko, ed. Handbook of augmented reality. Springer Science & Business Media, 2011.
K. D. Gupta, M. Ahsan, S. Andrei - Extending the Storage Capacity And Noise Reduction of a Faster QR-Code
71
Hartley, Richard, and Andrew Zisserman. „Multiple view geometry in computer vision”.
Cambridge University Press, 2003.
Liu, Yue, Ju Yang, and Mingjun Liu. (2008). "Recognition of QR Code with mobile phones".In
Control and Decision Conference, 2008. CCDC 2008. Chinese, pp. 203-206. IEEE.
Ouaviani, E., A. Pavan, M. Bottazzi, E. Brunelli, F. Caselli, and M. Guerrero. (1999)"A common
image processing framework for 2D barcode reading.",pp. 652-655.
Yang, Zhibo, Huanle Xu, Jianyuan Deng, Chen Change Loy, and Wing Cheong Lau. "Robust and
Fast Decoding of High-Capacity Color QR Codes for Mobile Applications". arXiv preprint
arXiv:1704.06447 (2017).
Ştefan ANDREI received his BSc in Computer Science (1994) and MSc in Computer Science
(1995) from Alexandru Ioan Cuza University of Iasi, and PhD in Computer Science (2000) from
Hamburg University. He was awarded with four competitive scholarships, as follows: the Singapore
- MIT Alliance Computer Science Fellowship, 2002-2005, the World Bank
Joint Japan Graduate Scholarship Program, 1998-2000, the TEMPUS
Fellowship, 1998-1998, and DAAD Scholarship, German Government, 5/1997-
7/1997. Dr. Andrei wrote over 100 publications published in prestigious
journals and conference proceedings which have more than 200 non-self
international citations in reputable publications. He has given invited talks at
more than 20 reputable universities and industrial companies. He has been a
Program Committee member or co-Chair of more than 40 international
reputable conferences and a PI, co-PI, or Senior Personnel of more than 11 funded research grant
proposals. He was promoted to ACM Senior Member in April 2013. Currently, he is an Associate
Professor and Chair of the Department of Computer Science with Lamar University, Beaumont,
TX, U.S.A. His research interests are in the areas of optimization techniques, verification, and
scheduling analysis for multiprocessor platforms for real-time embedded systems, software
engineering and translation systems.
