ROBUST INVISIBLE QR CODE IMAGE WATERMARKING IN DWT …€¦ · ROBUST INVISIBLE QR CODE IMAGE...
Transcript of ROBUST INVISIBLE QR CODE IMAGE WATERMARKING IN DWT …€¦ · ROBUST INVISIBLE QR CODE IMAGE...
International Journal of Electronics and Communication Engineering & Technology (IJECET),
ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 4, Issue 7 (2013) © IAEME
International Conference on Communication Systems (ICCS-2013) October 18-20, 2013
B K Birla Institute of Engineering & Technology (BKBIET), Pilani, India Page 190
ROBUST INVISIBLE QR CODE IMAGE WATERMARKING IN DWT
DOMAIN
V M Jithin1, K K Gupta2
EEE, BITS, Pilani, India
[email protected], [email protected]
ABSTRACT: Quick Response (QR) code is 2d code and widely used in magazine advertisement,
product packet, museum, and tour tickets. It has high data capacity compare to 1d code. The
invisible QR code is watermarked via popular wavelet transform algorithm in images. The
results show that the proposed method is robust and tested against attacks.
KEYWORDS: QR code, watermarking, Wavelets.
I. INTRODUCTION
In this modern era mobile and wireless technology is taking over all fields of life. They are
more than a communication device; they act as your PC, music player, your bank, your
shopping area and more [1]. In near future the smart glasses like google glass are going to
replace the current hand held devices which make to overlay the physical world with a digital
layer of tags, ads, maps etc [2].
In current smart phone scenario, one of the important technology used to connect the physical
world to internet or digital world using smart phone, is QR codes. They are used mainly for
marketing and commerce but their applications are numerous including virtual marketing,
virtual maps, QRpedia etc. We are making the QR code as invisible watermark in image using
digital watermarking technology. Other alternatives proposed were marking QR code in
invisible ink which is only visible with Ultraviolet [3], QR transparent stickers etc.
The main challenge in making invisible QR code is the detector part. Since the detection is
expected to be done in smart phone which has slower processors, the detection must be very
simple with very little processing required before passing information to QR reader. Here we
implement and simulate the QR encoder and decoder in MATLAB to study the effectiveness of
algorithm. This paper uses watermarking scheme in wavelet domain where multiple copies of
QR code is inserted into low frequency components of host image. Since the data requires no
secrecy compared to existing watermarking schemes, public watermarking technique is used
[4] [5] [6].
INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET )
ISSN 0976 – 6464(Print) ISSN 0976 – 6472(Online) Volume 4, Issue 7 (2013), pp. 190-195 © IAEME: www.iaeme.com/ijecet.asp Journal Impact Factor (2013): 5.8896 (Calculated by GISI) www.jifactor.com
IJECET © I A E M E
International Journal of Electronics and Communication Engineering & Technology (IJECET),
ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 4, Issue 7 (2013) © IAEME
International Conference on Communication Systems (ICCS-2013) October 18-20, 2013
B K Birla Institute of Engineering & Technology (BKBIET), Pilani, India Page 191
The watermarking is intended to encode secret or copyright information into host digital data
to demonstrate and protect the ownership of products. But in this paper we use this
technology to hide QR code in the background rather than using it to protect the information.
Anyway the same methodologies used in this paper can be used for copyright protection
process. In this paper invisible watermarking of QR code is done in wavelet domain.
The main advantages of invisible QR Code are 1. As opposed to adding yet another element to
the marketing piece or direct mailer, marketers can use existing images within the material to
use as a code; 2. It won’t waste advertisement area with QR.
This paper is organized as follows: QR code is explained in section 2. The implementation is
explained in section 3. The simulation and result are analyzed in section 4 and it followed by
conclusion.
II. QR CODE
QR Code is the trademark for a type of matrix barcode, first designed for the automotive
industry [7]. Two-dimensional bar code technology comparing with the traditional one-
dimensional bar code has the following advantages: 1. higher information density; 2. it can
express Chinese characters, images and even sound; 3.with error correction function. More
recently, the system has become popular outside the industry due to its fast readability and
large storage capacity compared to standard UPC barcodes. The code consists of black modules
(square dots) arranged in a square pattern on a white background. QR Code is capable of
handling all types of data, such as numeric and alphabetic characters, Kanji, Kana, Hiragana,
symbols, binary, and control codes. Up to 7,089 characters can be encoded in one symbol [6].
Data can be restored even if the symbol is partially dirty or damaged. A maximum 30% of
codewords can be restored. Fig. 1 shows the QR code structure.
Fig. 1: QR code structure
III. IMPLEMENTATION
The invisible watermarking of QR code is done in wavelet domain as shown in Fig. 2. Discrete
wavelet transform (DWT) is used due to its spatial resolution: it captures both frequency and
location information (spatial information). The multilevel DWT decomposition is used (4 level
Daubechies-4 wavelet transform). The block diagram of watermarking is shown in Fig. 2.
International Journal of Electronics and Communication Engi
ISSN 0976 – 6464(Print), ISSN 0976
International Conference on Communication Systems (ICCS
B K Birla Institute of Engineering & Technology (BKBIET), Pilani, India
The watermark is embedded into host image in 4th level of wavelet decomposition. As we go
deeper with decomposing low frequency sub image (approximate component) the data will get
more concentrate on approximate component in next levels. That is information in high
frequency components (horizontal, vertical and diagonal) will be very less compared to low
frequency component. So inserting can be done on high frequency component since it won’t
destroy high amount of information. Usually these components will be black with small amount
of white patches which corresponds to high frequency portion (sudden variation in brightness)
of host image. So the location of insertion of watermark should be suitably sele
won’t include sudden variations. In this paper we use multiple watermarking techniques
where three copies of QR code is inserted in all three high frequency sub images after multiple
levels of wavelet decomposition. The location of insertio
subimages since at time of reception it would be easier in slower processors to recover QR
code easily by simple pixel manipulations like addition or correlation. The location is found
using approximate subimage such that i
minimum variation in pixel values.
Another important thing to be noted while insertion is the peak level of pixel value of QR code.
Our empirical studies showed that the insertion with pixel value greater than one by fifteenth
of mean value of approximate subimage give better results while if value exceeds one by
hundredth of maximum pixel value of approximate image make the QR visible after
reconstruction. So trade off done between these values to set the maximum pixel value of
inserted watermark.
International Journal of Electronics and Communication Engineering & Technology (IJECET),
6464(Print), ISSN 0976 – 6472(Online), Volume 4, Issue 7 (2013)
International Conference on Communication Systems (ICCS-2013)
of Engineering & Technology (BKBIET), Pilani, India
The watermark is embedded into host image in 4th level of wavelet decomposition. As we go
deeper with decomposing low frequency sub image (approximate component) the data will get
pproximate component in next levels. That is information in high
frequency components (horizontal, vertical and diagonal) will be very less compared to low
frequency component. So inserting can be done on high frequency component since it won’t
h amount of information. Usually these components will be black with small amount
of white patches which corresponds to high frequency portion (sudden variation in brightness)
of host image. So the location of insertion of watermark should be suitably sele
won’t include sudden variations. In this paper we use multiple watermarking techniques
where three copies of QR code is inserted in all three high frequency sub images after multiple
levels of wavelet decomposition. The location of insertion is made same for all the three
subimages since at time of reception it would be easier in slower processors to recover QR
code easily by simple pixel manipulations like addition or correlation. The location is found
using approximate subimage such that in particular block selected for insertion is the one with
minimum variation in pixel values.
Fig. 2: watermarking procedure
Another important thing to be noted while insertion is the peak level of pixel value of QR code.
that the insertion with pixel value greater than one by fifteenth
of mean value of approximate subimage give better results while if value exceeds one by
hundredth of maximum pixel value of approximate image make the QR visible after
ade off done between these values to set the maximum pixel value of
Fig. 3: recovery of watermark
neering & Technology (IJECET),
Volume 4, Issue 7 (2013) © IAEME
October 18-20, 2013
Page 192
The watermark is embedded into host image in 4th level of wavelet decomposition. As we go
deeper with decomposing low frequency sub image (approximate component) the data will get
pproximate component in next levels. That is information in high
frequency components (horizontal, vertical and diagonal) will be very less compared to low
frequency component. So inserting can be done on high frequency component since it won’t
h amount of information. Usually these components will be black with small amount
of white patches which corresponds to high frequency portion (sudden variation in brightness)
of host image. So the location of insertion of watermark should be suitably selected so that it
won’t include sudden variations. In this paper we use multiple watermarking techniques
where three copies of QR code is inserted in all three high frequency sub images after multiple
n is made same for all the three
subimages since at time of reception it would be easier in slower processors to recover QR
code easily by simple pixel manipulations like addition or correlation. The location is found
n particular block selected for insertion is the one with
Another important thing to be noted while insertion is the peak level of pixel value of QR code.
that the insertion with pixel value greater than one by fifteenth
of mean value of approximate subimage give better results while if value exceeds one by
hundredth of maximum pixel value of approximate image make the QR visible after
ade off done between these values to set the maximum pixel value of
International Journal of Electronics and Communication Engineering & Technology (IJECET),
ISSN 0976 – 6464(Print), ISSN 0976 – 6472(Online), Volume 4, Issue 7 (2013) © IAEME
International Conference on Communication Systems (ICCS-2013) October 18-20, 2013
B K Birla Institute of Engineering & Technology (BKBIET), Pilani, India Page 193
At decomposition phase most simple algorithm is used to decrease computational complexity.
We use blind recovery technique which does not require the original image for recovery. We
use QR decoder developed by Z Xing (Zebra barcode decoder an open-source, multi-format
1D/2D barcode image processing library implemented in Java, with ports to other languages
powered by google code project). First we take multilevel DWT and give each high frequency
subimage directly to QR decoder. In the absence of any noise it gives direct output. If QR
decoder fails to decode it directly use group of subimages and use majority check to remove
pixel errors and again give it to QR decoder. Still the QR coder fails which happens in presence
of high noise we use threshold techniques to get a binary image with minimum bit errors
compared with original QR. Our studies show that thresholding at levels between 0.3 and 0.4
gives better results. Those data are provided in simulation and analysis section. The block
diagram for recovery algorithm is shown in Fig. 3.
IV. SIMULATION AND ANALYSIS
The entire simulation is done using Matlab. The algorithm is tried for more than twenty five
images and the study mainly concentrated on 4 test images – Lena, Baboon, Air Plane and
House Boat. The first three are 512x512 image while last one is 960x536 pixels. QR code for
testing is generated using QR code generator by Z. Xing project. While the maily used QR is
33x33 pixel QR (encoded with www.facebook.com). The Peak signal to noise ratio (PSNR) is
measured after embedding invisible QR code. The PSNR is given in Table 1. The qualitative
result is shown in Fig. 4
Test Image Lena Baboon Airplane Houseboat
Size 512x512 512x512 512x512 960x536
PSNR 72.66 73.0749 66.5892 73.1276
Table 1: Output performance is measures using peak signal to noise ratio (PSNR) calculation
Fig. 4: The qualitative results- first image without and next same image with watermark
To find out robustness of proposed watermarking techniques, the efficiency is checked with
various attacks. In our smart phone camera scenario most common attacks on images are noise
due to grain, JPEG compression, cropping or white spaces at outer parts of images and rotation.
We tried to model these attacks and found out robustness against them.
International Journal of Electronics and Communication Engi
ISSN 0976 – 6464(Print), ISSN 0976
International Conference on Communication Systems (ICCS
B K Birla Institute of Engineering & Technology (BKBIET), Pilani, India
Fig. 5: The number of detection Vs
To analyze noise performance of decoding algorithm we used Lena test image with water mark
as base. The salt and pepper noise is added and plot between PSNR and no of
shown in Fig. 5. The algorithm woks well when PSNR is greater than 28 dB.
Fig.
To find out thresholding level noise is fixed at a given value and no of bit errors are finding
with thresholding level between
pepper noise at 0.05 level is shown in Fig
Fig. 7: Number of detection Vs JPEG compression
Fig. 7 shows watermarked image with varying level of JPEG compression. Since the watermark
is inserted in wavelet domain chance of distortion is more as extend of compression increases.
Simulating Lena image with various levels of JPEG compression shows that faithful detection is
not possible if compression is more than 60%. One of the highest p
watermark image is cropping. The chance of small parts at outer ridge of image be cropped
while taking photo in mobile camera is more. To avoid data loss due to these type of attack a
10% margin is set where watermark is not in
part is got cropped chance of recovery is little due to spatial properties of wavelet transform.
International Journal of Electronics and Communication Engineering & Technology (IJECET),
6464(Print), ISSN 0976 – 6472(Online), Volume 4, Issue 7 (2013)
International Conference on Communication Systems (ICCS-2013)
of Engineering & Technology (BKBIET), Pilani, India
The number of detection Vs PSNR with salt and pepper noise
To analyze noise performance of decoding algorithm we used Lena test image with water mark
as base. The salt and pepper noise is added and plot between PSNR and no of
5. The algorithm woks well when PSNR is greater than 28 dB.
Fig. 6: Bit error Vs threshold value
To find out thresholding level noise is fixed at a given value and no of bit errors are finding
with thresholding level between 0 and 1. The plot for bit error vs. threshold value for salt and
pepper noise at 0.05 level is shown in Fig. 6.
Number of detection Vs JPEG compression
7 shows watermarked image with varying level of JPEG compression. Since the watermark
is inserted in wavelet domain chance of distortion is more as extend of compression increases.
Simulating Lena image with various levels of JPEG compression shows that faithful detection is
not possible if compression is more than 60%. One of the highest probable chances of attack on
watermark image is cropping. The chance of small parts at outer ridge of image be cropped
while taking photo in mobile camera is more. To avoid data loss due to these type of attack a
10% margin is set where watermark is not inserted. Even though for smaller images if large
part is got cropped chance of recovery is little due to spatial properties of wavelet transform.
neering & Technology (IJECET),
Volume 4, Issue 7 (2013) © IAEME
October 18-20, 2013
Page 194
PSNR with salt and pepper noise
To analyze noise performance of decoding algorithm we used Lena test image with water mark
as base. The salt and pepper noise is added and plot between PSNR and no of detection as
5. The algorithm woks well when PSNR is greater than 28 dB.
To find out thresholding level noise is fixed at a given value and no of bit errors are finding
0 and 1. The plot for bit error vs. threshold value for salt and
7 shows watermarked image with varying level of JPEG compression. Since the watermark
is inserted in wavelet domain chance of distortion is more as extend of compression increases.
Simulating Lena image with various levels of JPEG compression shows that faithful detection is
robable chances of attack on
watermark image is cropping. The chance of small parts at outer ridge of image be cropped
while taking photo in mobile camera is more. To avoid data loss due to these type of attack a
serted. Even though for smaller images if large
part is got cropped chance of recovery is little due to spatial properties of wavelet transform.
International Journal of Electronics and Communication Engi
ISSN 0976 – 6464(Print), ISSN 0976
International Conference on Communication Systems (ICCS
B K Birla Institute of Engineering & Technology (BKBIET), Pilani, India
V. CONCLUSION AND FUTURE
In this paper we try to model a new scheme to embed QR code in host images so as to
hidden from visual perception. We used multilevel wavelet transform and multiple copies
watermark is inserted for faithful and easy detection. The insertion location and value are
found using algorithm. Also we checked the robustness of technique
results show that this scheme gives better result even in presence of noise (When PSNR greater
than 28dB). It can resist JPEG compression upto 60%. Data can be recovered when image is
cropped along the boundary region. However pe
smartphone environment in this study. We are looking into the implementation of this
algorithm in android and iphone platforms and further modification based on performance.
REFERENCES
[1]Lee Garber, Scanning the Future with New Barcodes, IEEE Computer Magazine, 2011, 20
[2]Chung-Hsin Liu Chia-Hong Chou, Two
Implementation and Performance Analysis, IEEE International Conference, 632
[3]K. Kamijo, N. Kamijo, Zhang Gang, Invisible barcode with optimized error correction, IEEE
International Conference on Image Processing, 2008. ICIP 2008, 2036
[4]P.H.W. Wong, O.C. Au, and Y.M. Yeung, A novel blind multiple watermarking technique for
images, IEEE Trans. Circuits System Video Technology, 13(), 2003, 813
[5]P. Premaratne, C.C. Ko, A novel watermark embedding and detection scheme for images in
DFT domain, Proc. 7th Int. IPA, 2, 1999, 780
[6]C. T. Hsu, and J. L. Wu, Hidden digital wate
Processing, 8(), 1999, 58-68.
[7]QR code website -- http://www.qrcode.com/en/index.html
[8]C. S. Lu, and H. Y. M. Liao, Multipurpose watermarking for image authentication and
protection, IEEE Trans. Image Proces
BIOGRAPHY
Jithin V M was born in
B.Tech degree in Electronics and
Government
pursuing his M.
Rajasthan,India. His current research interests focus on
DSP and Wireless communication
Gupta K K was born in
BITS, Pilani, India in
Processing, DSP and
International Journal of Electronics and Communication Engineering & Technology (IJECET),
6464(Print), ISSN 0976 – 6472(Online), Volume 4, Issue 7 (2013)
International Conference on Communication Systems (ICCS-2013)
of Engineering & Technology (BKBIET), Pilani, India
UTURE WORK
In this paper we try to model a new scheme to embed QR code in host images so as to
hidden from visual perception. We used multilevel wavelet transform and multiple copies
watermark is inserted for faithful and easy detection. The insertion location and value are
found using algorithm. Also we checked the robustness of technique with various attacks. The
that this scheme gives better result even in presence of noise (When PSNR greater
than 28dB). It can resist JPEG compression upto 60%. Data can be recovered when image is
cropped along the boundary region. However performance of this algorithm is not tested in
smartphone environment in this study. We are looking into the implementation of this
algorithm in android and iphone platforms and further modification based on performance.
[1]Lee Garber, Scanning the Future with New Barcodes, IEEE Computer Magazine, 2011, 20
Hong Chou, Two-dimensional bar code mobile commerce
Implementation and Performance Analysis, IEEE International Conference, 632
mijo, N. Kamijo, Zhang Gang, Invisible barcode with optimized error correction, IEEE
International Conference on Image Processing, 2008. ICIP 2008, 2036-2039.
[4]P.H.W. Wong, O.C. Au, and Y.M. Yeung, A novel blind multiple watermarking technique for
, IEEE Trans. Circuits System Video Technology, 13(), 2003, 813-830.
[5]P. Premaratne, C.C. Ko, A novel watermark embedding and detection scheme for images in
DFT domain, Proc. 7th Int. IPA, 2, 1999, 780-783.
[6]C. T. Hsu, and J. L. Wu, Hidden digital watermarks in images, IEEE Transactions on Image
http://www.qrcode.com/en/index.html
[8]C. S. Lu, and H. Y. M. Liao, Multipurpose watermarking for image authentication and
protection, IEEE Trans. Image Processing, 10(), 2001, 1579-1592.
was born in Payyanur, Kerala, India in 1989. He received the
B.Tech degree in Electronics and Communication Engineering from
College of Engineering Kannur (Kerala), India in 2011. He is
pursuing his M. E in Communication Engineering
India. His current research interests focus on
Wireless communication.
was born in UP, India in 1969. He received t
India in 2008. His current research interests focus on
, DSP and Instrumentation.
neering & Technology (IJECET),
Volume 4, Issue 7 (2013) © IAEME
October 18-20, 2013
Page 195
In this paper we try to model a new scheme to embed QR code in host images so as to make it
hidden from visual perception. We used multilevel wavelet transform and multiple copies
watermark is inserted for faithful and easy detection. The insertion location and value are
with various attacks. The
that this scheme gives better result even in presence of noise (When PSNR greater
than 28dB). It can resist JPEG compression upto 60%. Data can be recovered when image is
rformance of this algorithm is not tested in
smartphone environment in this study. We are looking into the implementation of this
algorithm in android and iphone platforms and further modification based on performance.
[1]Lee Garber, Scanning the Future with New Barcodes, IEEE Computer Magazine, 2011, 20-21.
dimensional bar code mobile commerce
Implementation and Performance Analysis, IEEE International Conference, 632-635.
mijo, N. Kamijo, Zhang Gang, Invisible barcode with optimized error correction, IEEE
2039.
[4]P.H.W. Wong, O.C. Au, and Y.M. Yeung, A novel blind multiple watermarking technique for
830.
[5]P. Premaratne, C.C. Ko, A novel watermark embedding and detection scheme for images in
rmarks in images, IEEE Transactions on Image
[8]C. S. Lu, and H. Y. M. Liao, Multipurpose watermarking for image authentication and
, India in 1989. He received the
ommunication Engineering from
), India in 2011. He is
at BITS Pilani,Pilani
India. His current research interests focus on Image Processing,
9. He received the PhD degree from
. His current research interests focus on Image