VLSI IMPLEMENTATION OF EFFICIENT IMAGE

WATERMARKING ALGORITHM
SURAJ KUMAR DHANUKA PRIYAM SACHDEVA SHARUKH S SHAIKH
VIT UNIVERSITY VIT UNIVERSITY NITK
CHENNAI CHENNAI SURATKAL
[email protected] [email protected] [email protected]

SHASHANK ALEVOOR
NITK
SURATKAL
[email protected]

Abstract— Protection and authentication of digital multimedia domain transformations to gain the benefits of both. The
content during transmission has become very important with the approach is to embed the bit plane sliced watermark into a bit
ongoing development in communication and networking field. plane sliced mid-high frequency component of discretely
Aiming at this popular research topic of recent time, this paper wavelet transformed host image.
presents a comparative analysis of different watermarking
The paper is organized in a structured manner as followed:
techniques performed in MATLAB. It also presents a robust
VLSI architecture for watermarking which survives most of the
a brief Introduction about the topics related to paper in Section
attacks to which data is exposed on internet and contributes II with a compendium review of related works. Section III
towards anti-plagiarism rule. This paper exhibits the ASIC presents a detailed description of proposed method. Section IV
implementation of an efficient digital watermarking algorithm reports and describes the experimental results and observations.
for images based on frequency domain Discrete Wavelet Section V emphasizes on the conclusion of the paper.
Transform and spatial domain bit plane slicing.
II. LITERATURE REVIEW
Index Terms—Image Watermarking, Discrete Cosine Transform
(DCT), Discrete Wavelet Transform (DWT), Bit plane slicing, Very In recent time, many researchers are trying to develop a
Large Scale Integration (VLSI), Application Specific Integrated memory, area and power efficient hardware implementation of
Circuit (ASIC). watermarking algorithm. To increase efficiency of
watermarking, the frequency domain transformation used
should be optimized first. Thus, the authors of [1] presented an
I. INTRODUCTION efficient architecture in terms of memory for lifting based
Digital Image Watermarking has emerged as an effective DWT supporting dual mode: 9/7 lossy and 5/3 lossless coding
and trustworthy solution as a proof of one’s ownership and showing remarkable reduction in latency and transpose
copyright protection even in the cases where the protected data memory with regular signal flow. The researchers in [2]
is attacked in different possible manners. proposed a new frequency domain DWT based watermarking
algorithm having an additional watermark generation stage
Watermarking procedure consists of four stages: structuring
where watermark was constructed using host image and was
watermark, watermark embedding, processing watermarked
disordered using Arnold transform. The writers of [7] reviewed
image, extracting watermark. Watermarking for digital images various watermarking techniques with special focus on
can be categorized into frequency domain watermarking or hardware implementation and presented the differences and
spatial domain watermarking. The most frequently used similarities between these different types based on their
frequency domain transformations are DCT and DWT. The assessment. The experimenters in [8] proposed an effective
well-known and widely used spatial domain methods are least blind digital watermarking algorithm where the original image
significant bit (LSB) and histogram based. Digital is DWT transformed and using a secret key a robust
watermarking performed in spatial-domain consists of easy watermarked image is generated.
operations and are generally computational cost efficient but
gives poor results after common signal processing like III. IMPLEMENTATION
filtering and JPEG etc. and also after geometric distortions
like scaling and cropping etc. Whereas, watermarking The paper implements and compares the result of lifting
techniques in frequency domain proves to be greatly robust scheme based Haar discrete wavelet transform and bit plane
slicing based digital image watermarking approach with the
against different attacks but consists of computational
previously implemented watermarking approaches. The
complexities. algorithm used for implementing this proposed watermarking
Thus, in this paper we have opted for a mingled approach for approach is clearly explained below using the flowcharts.
digital image watermarking using spatial and frequency

978-1-4799-8047-5/15/$31.00 2015
c IEEE 1002
 The DWT and IDWT approach followed is based on Haar The flowchart depicting the procedure for embedding
wavelet transform which uses the following equations: watermark in host image during image watermarking process
using HL coefficients is shown in Fig 1 and same is followed
The equations used for determining the approximation for LH embedding. In this process the original gray scale
(high) and detailed (low) component during forward transform image is transformed using frequency domain DWT to obtain
and even and odd components (pixels) of the image during the HH, HL, LL, LH coefficients and then watermark is inserted in
inverse transform are: the suitable coefficients of transformed host image using
…….eq. 1 spatial domain bit plane slicing. Then it is again back
transformed or inversely transformed to generate watermarked
…….eq. 2 image.
……….eq. 3 The HH and LL quadrants are not used for embedding
watermark as they do not provide a robust and reliable solution
…….eq. 4 and also the extraction of watermark from watermarked image
The paper implements the algorithm in three stages: is not possible under various circumstances. This paper draws a
comparative analysis based on image quality measures for
Stage 1: This watermark embedding stage consists of watermark embedded in HL and LH quadrants.
generating watermarked image from the host image and
watermark image with the help of Haar based DWT and BPS In extraction process the watermarked image is taken and
transformed using DWT Haar equations mentioned previously
Stage 2: The processing watermarked image stage where and the coefficients for HH, HL, LH and LL are obtained.
the watermarked image is attacked to check for its robustness Using the pixel values of quadrant LH, the lower 4 bits of each
under different extreme conditions prevailing over transmission pixel is extracted and used as higher 4 bits of pixel values of
media. watermark image for the formation of extracted watermark
Stage 3: The watermark extraction stage consists of image.
extracting watermark image from watermarked image without The extraction process used in this technique is blind, that
the usage of original image. is during extraction the original image is not used to obtain the
START
watermark from the watermarked image.

Original Image IV. EXPERIMENTAL RESULTS

The paper implements different approaches for digital image
Divide Image into Even & watermarking in MATLAB and compares the results based on
Odd Components
image quality parameters such as Peak Signal To Noise Ratio
(PSNR), Mean Square Error (MSE), Structural Content (SC),
From Even & Odd Values Average Difference (AD), Normalized Cross Correlation
Calculate High & Low
Components Using Eqn. 1, 2 (NCC), Normalized Absolute Error (NAE), Maximum
Difference (MD), and Processing Time (PT). The original
image taken is Lena image of size 256x256 and watermark
used is a ‘CS’ logo of size 128x128.
Again Divide High Coefficients Again Divide Low Coefficients
into Even & Odd Components into Even & Odd Components The Table 1 presents the comparative analysis of different
watermarking techniques based on the image quality measures
when watermark is embedded in HL sub-band.
Using Even & Odd Values
The Table 2 presents the comparative analysis of different
Using Even & Odd Values
Calculate High And Low Calculate High And Low watermarking techniques based on the image quality measures
Components Using Eqn. 1, 2 Components Using Eqn. 1, 2 when watermark is inserted in LH sub-band.
Based on the analysis of these tabulated results, it can be
HL HH LH LL
clearly derived that the watermarking technique using a
combination of HAAR based DWT and bit plane slicing is
more efficient and effective watermarking technique.
Using Bit Plane Slicing Insert
Higher 4 Bits of Watermark in
Thus, this technique is implemented in Verilog followed by its
Watermark Image
Lower 4 Bits of Hl ASIC implementation.
Figure 2 depicts the results obtained with this technique after
embedding the watermark and after extracting the watermark
Perform IDWT Using HH, LL, LH from Verilog.
& New HL With Eqn. 3, 4

Watermarked Image

Fig. 1: Watermarked image, Original Watermark, Extracted Watermark

2015 IEEE International Advance Computing Conference (IACC) 1003

 (e.)
Fig. 3: Results after various attacks, (a) watermarked image after Gaussian
Fig. 2: Watermarked image, Original Watermark, Extracted Watermark
noise attack and extracted watermark, (b) watermarked image after Salt &
Pepper noise and extracted watermark, (c) watermarked image after Speckle
The watermarked image was then attacked with different noise and extracted watermark, (d) watermarked image after Blurring and
degrading factors like Gaussian noise, speckle noise, salt and extracted watermark, (e) watermarked image after Cropping and extracted
pepper noise, scaling, cropping, blurring, intensity variation watermark.
and contrast enhancement.
Figure 3 shows the watermarked image and watermark The schematic diagram obtained using Cadence Encounter RC
extracted after the attack using Gaussian noise, Salt and pepper tool after the complete simulation of the Verilog code for the
noise, Speckle noise, Blurring & cropping attack. image watermarking technique proposed in this paper is
shown in the following Fig. 4.

(a.)

Fig. 4 Schematic for Watermarking Block

(b.) From the timing report for this watermarking technique
generated in Cadence Encounter RC, we can infer that the
slack is 186ps, rise slew is 26ps and fall slew is 27ps.

Table 3: ASIC implementation report for Area and Power

Cells 26290
Cell Area 44637
(c.) Net Area 83794
Total Area 128431
Leakage Power (nW) 2200.380
Internal Power (nW) 1415919.64
Net Power (nW) 1113880.28
Switching Power (nW) 302039.36

(d.)

Table 1: Comparison of Image Quality Measure for different watermarking techniques when embedded in HL sub band
HL DCT1 [9] DCT2 [10] DWT DWT- DWT-DWT DWT- DWT-DCT DWT BIT DWT DWT HAAR
[11] DWT [3] BIT PLANE DCT [4] BIT PLANE PLANE HAAR [6] BIT PLANE
[12] [13] [5]
MSE 0.2051 0.2051 0.3460 0.3226 1.1116 1.4930 0.6052 0.3870 0.1196 0.0674
PSNR 54.8745 54.8745 52.7399 53.0446 47.6712 46.3902 50.3118 52.2534 57.3522 59.84181
NCC 1.0000 1.0000 0.9985 0.9986 0.9963 0.9982 1.0009 0.9988 0.9999 1.000
AD -0.0052 -0.0014 0.1967 -2.4593 -2.1340 -2.4094 -2.7882 -2.4866 0.0138 0.0013
SC 0.9999 1.0000 0.9855 0.9853 0.9897 0.9860 0.9808 0.9849 1.0002 1.000
MD 16.4509 40.0801 11 11 13 26 16 11 1 1
NAE 0.00041847 0.00012468 0.0224 0.0223 0.0256 0.0233 0.0247 0.0228 9.8198e-4 5.4102e-4
PSNR W 283.2118 328.4825 infinity Infinity Infinity 20.3972 15.2082 infinity 69.8057 Infinity
PT(s) 3.574562 2.227043 4.958180 6.820727 5.931720 6.067988 5.257976 5.363030 4.248876 3.577177

1004 2015 IEEE International Advance Computing Conference (IACC)

Table 2: Comparison of Image Quality Measure for different watermarking techniques when embedded in LH sub band
LH DCT1 [9] DCT2 [10] DWT DWT- DWT-DWT DWT- DWT-DCT DWT DWT DWT HAAR
[11] DWT [3] BITPLANE DCT [4] BITPLANE BITPLANE HAAR [6] BITPLANE
[12] [13] [5]

MSE 0.2051 0.2051 0.1036 0.1070 0.1461 0.1165 1.6550 0.1317 0.1165 0.0673

PSNR 54.8745 54.8745 57.9765 57.8362 56.4857 57.4676 45.9429 56.9337 57.4683 59.8497
NCC 1.0000 1.0000 1.0005 1.0004 1.00037 1.0011 0.9982 1.0008 1.000 1.000
AD -0.0052 -0.0014 0.0605 -2.7110 -3.2045 -2.8032 -2.7734 0.0036 0.0047
-2.4116
SC 0.9999 1.0000 0.9817 0.9818 0.9751 0.9804 0.9860 0.9809 1.0000 1.000
MD 16.4509 40.0801 11 11 11 11 20 11 1 1
NAE 4.1847e-4 1.2468e-4 0.0204 0.224 0.0265 0.0232 0.0233 0.0231 9.5668e-4 5.4005e-4
PSNR W 283.2118 328.4825 Infinity Infinity Infinity 20.3972 18.1338 infinity 76.9914 infinity
PT(s) 3.038622 2.182874 5.098104 6.406768 5.314431 5.496333 5.057976 6.003690 3.333995 3.187528

Table 4: PSNR and correlation after attack [2] Mohamed Sathik, SS Sujatha, “A novel DWT based Invisible
Attacks PSNR PSNR Cross Watermarking Technique for Digital Images”, International Arab
(Image) (Watermark) Correlation Journal of e-Technology, Jan 2, Vol.2, No.3
[3] Anumol T J, Binson V A, Soumya Rasheed, “FPGA Implementation
Gaussian Noise 57.4279 58.5643 0.9113 of Low Power, High Speed, Area Efficient Invisible Image
Salt & Pepper 15.8811 56.0383 0.8400 Watermarking Algorithm for Images”, International Journal Of
Noise Scientific & Engineering Research Volume 4, August-2013, Issue 8
[4] Li Sukang, Mei Jiansheng, and Tan Xiaomei, “Digital Watermarking
Speckle Noise 55.1123 56.2556 0.8445 Algorithm Based On DCT and DWT”, Proceedings of the 2009
Attack Blur 55.9764 57.7885 0.8850 International Symposium on Web Information Systems and
Applications (WISA’09), China, May 22-24, 2009, pp. 104-107
[5] K.Esther Rani, S.Bhargav Kumar, “FPGA Implementation of 4-D
The result of the ASIC implementation of DWT BPS based DWT and BPS based Digital Image Watermarking”, International
image watermarking using Cadence RC tool is shown in Journal of Engineering Trends and Technology- Volume3, Issue2-
Table 3. Table 4 tabulates the PSNR of attacked 2012
watermarked image and PSNR and correlation of attacked [6] D.S. Bormane, Pallavi Patil, “DWT Based Invisible Watermarking
Technique for Digital Images”, International Journal of Engineering
extracted watermark. and Advanced Technology (IJEAT) , April 2013, ISSN: 2249 – 8958,
Volume-2, Issue-4
V. CONCLUSION [7] Rameshwar Rao, Mustafa Osman Ali, “An Overview of Hardware
The paper presents a comparative analysis of various Implementation for Digital Image Watermarking”, 2011 International
Conference on Signal, Image Processing and Applications With
techniques for digital image watermarking implemented in workshop of ICEEA 201, IPCSIT vol.21 (2011), IACSIT Press,
MATLAB and compared against standard Image Quality Singapore
Parameters. Analyzing the results obtained from this study it [8] Ikpyo Hong, Intaek Kim, Seung-So0 Han, “A Blind Watermarking
was concluded that the approach based on amalgamation of Technique Using Wavelet Transform”, ISIE 2001, Pusan, KOREA, 0-
Haar DWT and bit plane slicing is better in terms of 7803-70~9-2/01, IEEE
transparency, robustness, unambiguous and payload. The [9] Pravin M. Pithiya, H.L.Desai, “DCT Based Digital Image
paper also proposes and implements an efficient digital Watermarking, Dewatermarking & Authentication”, International
image watermarking architecture using a combination of Journal of Latest Trends in Engineering and Technology (IJLTET) ,
Issue 3 May 2013, Vol. 2
spatial domain bit plane slicing and frequency domain Haar
[10] Shahin Shaikh, Manjusha Deshmukh, “Digital Image Watermarking
based DWT in VLSI. The watermarked image is also In DCT Domain, International Journal of Emerging Technology and
checked for ease of extraction after getting attacked by Advanced Engineering, Volume 3, Issue 4, April 2013
various attacks like addition of noise, contrast enhancement, [11] Shikha Tripathy, RC Jain, “Anovel DCT And DWT Based
blurring and resizing. Watermarking Techniques For Digital Images”, 18th international
conference on pattern recognition, 2006, IEEE Computer Society
[12] BN Chatterjee, “A New Wavelet Based Logo Watermarking
REFERENCES Scheme”, Elsevier 2004, doi:10.1016/j.patrec.2004.09.047
[1] Chih-Hsien Hsia, Jing-Ming Guo, Jen-Shiun Chiang, “Memory- [13] V. Ashok Kumar, J. Sateesh, P. Sirish Kumar, “Effective Digital
Efficient Hardware Architecture of 2-D Dual-Mode Lifting-Based Watermarking Approach Using Combined DWT DCT”, International
Discrete Wavelet Transform” , IEEE Transactions On Circuits And Journal of Engineering Research and Technology, 2013
Systems For Video Technology, April 2013, Vol. 23, No. 4

2015 IEEE International Advance Computing Conference (IACC) 1005