A fragile watermarking based on binomial transform in color images

Jan Zizka (Eds) : CCSIT, SIPP, AISC, PDCTA - 2013
pp. 281–288, 2013. © CS & IT-CSCP 2013 DOI : 10.5121/csit.2013.3632
A FRAGILE WATERMARKING BASED ON
BINOMIAL TRANSFORM IN COLOR
IMAGES
J. K. Mandal and S. K.Ghosal
Department of Computer Science and Engineering,
Kalyani University, Kalyani,
West Bengal, India, 741235.
{jkm.cse@gmail.com, sudipta.ghosal@gmail.com}
http://www.klyuniv.ac.in;http://www.jkmandal.com/
ABSTRACT
In this paper, a novel Binomial Transform based fragile image watermarking technique has
been proposed for color image authentication. The Binomial Transform (BT) is used to convert
each 2 × 2 sub-image block of the carrier image into transformed block corresponding to red,
green and blue channels in sliding window manner. One/two/three watermark bits are
embedded in second/third/fourth transformed components starting from the least significant
bit’s position (LSB-0). An adjustment has been incorporated to adjust embedded component
closer to the actual value without hampering the fabricated bits. The inverse Binomial
transform (IBT) is used to convert the transformed components back into the spatial domain. A
delicate re-adjustment method is applied on the first transformed component to remain the pixel
components in a valid range. The embedding operation in succession generates the final
watermarked image. At the receiving end, whole watermark is extracted based on the reverse
procedure and authentication is done through computed message digest and extracted bits.
Experimental results conform that the proposed technique produces high payload and Peak
Signal to Noise Ratio (PSNR) as compared to existing transformation based techniques[1, 5].
KEYWORDS
BT, IBT, LSB-0, Payload, PSNR, Authentication and Message digest.
1. INTRODUCTION
Digital watermarking is a technique of incorporating useful information into various digital media
like image, audio and video etc. for ownership evidence, fingerprinting, authentication and
integrity verification. The objective of this paper is to verify the integrity of a carrier image based
on a novel fragile watermarking technique in transform domain.
Various transformations are applied to convert the carrier image from spatial domain to transform
domain in a block wise manner. Each block contains a set of transformed components which can

282 Computer Science & Information Technology (CS & IT)
be slightly modified to embed the watermark data. The watermarked image can be obtained by
applying the respective inverse transformation. A separable discrete Hartley transform based
invisible watermarking scheme[1]
scheme exploits for image authenticating purpose by fabricating the authenticating watermark
data along with the message digest (which is generated from authenticating data) into the carrier
image with a minimal loss of quality and improved security.
The effectiveness of different watermarking methods can be measured in terms of payload, peak
signal to noise ratio and image fidelity etc. In this regard, the Binomial transformation [2
been applied for embedding watermark data into the cover image with a high payload and less
degradation in quality. Moreover, a 128 bit message digest has been used for authentication
purpose[4,5]. On embedding authenticating watermark (message/image) bits, inverse Bin
transformation is applied to re-generate the watermarked image in spatial domain.
The binomial transform (BT) is applied on pixel components {a
components {sn} as per equation (1).
Similarly, the inverse Binomial transform (IBT) is used to convert transformed components back
into pixel domain as per equation (2).
The main objective of proposed technique emphasizes on color image authentication by
protecting secret watermark. The message digest MD (which is generated from watermark data)
is used for authentication by verifying the integrity of the carrier image.
Section 2 of the paper deal with the proposed technique. Results, comparison and analysis are
given in section 3. Conclusions are drawn in section 4. References are given at end.
2. THE TECHNIQUE
In this paper, a novel color image authentication
transform. The message digests (MD) is obtained from the authenticating watermark
(message/image). The message digests and the size of the authenticating watermark is embedded
using the proposed technique as an
applied on each 2 × 2 sub-image block of the carrier image to convert the pixel components into
transformed components in row major order. One/two/three bits from the message digest,
watermark size and the watermark are embedded on second/third/fourth transformed components
starting from the least significant bits position (LSB
into each embedded component to reduce the changes made due to embedding. An invers
Binomial transform (IBT) has been applied on each adjusted component to get back the pixel
components. During the IBT, if the pixel component becomes negative or greater than 255, a re
Computer Science & Information Technology (CS & IT)
invisible watermarking scheme[1] has been proposed by Mandal & Ghoshal. The SDHTIWCIA
minimal loss of quality and improved security.
signal to noise ratio and image fidelity etc. In this regard, the Binomial transformation [2
for embedding watermark data into the cover image with a high payload and less
purpose[4,5]. On embedding authenticating watermark (message/image) bits, inverse Bin
generate the watermarked image in spatial domain.
The binomial transform (BT) is applied on pixel components {an} to generate transformed
} as per equation (1).
(1)
into pixel domain as per equation (2).
(2)
is used for authentication by verifying the integrity of the carrier image.
given in section 3. Conclusions are drawn in section 4. References are given at end.
In this paper, a novel color image authentication technique has been proposed based on Binomial
using the proposed technique as an initial embedding. The Binomial transform (BT) has been
image block of the carrier image to convert the pixel components into
nd the watermark are embedded on second/third/fourth transformed components
starting from the least significant bits position (LSB-0). An adjustment method has been applied
into each embedded component to reduce the changes made due to embedding. An invers
components. During the IBT, if the pixel component becomes negative or greater than 255, a re
has been proposed by Mandal & Ghoshal. The SDHTIWCIA
signal to noise ratio and image fidelity etc. In this regard, the Binomial transformation [2-3] has
for embedding watermark data into the cover image with a high payload and less
purpose[4,5]. On embedding authenticating watermark (message/image) bits, inverse Binomial
} to generate transformed
technique has been proposed based on Binomial
initial embedding. The Binomial transform (BT) has been
image block of the carrier image to convert the pixel components into
nd the watermark are embedded on second/third/fourth transformed components
0). An adjustment method has been applied
into each embedded component to reduce the changes made due to embedding. An inverse
components. During the IBT, if the pixel component becomes negative or greater than 255, a re-

Computer Science & Information Technology (CS & IT) 283
adjustment operation is to be performed on the first transformed component repeatedly, till the
pixel component become greater than or equal to 0 and less than or equal to 255. The procedure is
applied for red, green and blue channel separately and the process continues till the final
watermarked image is produced. The recipient of the watermarked image is instructed to perform
the reverse operation to extract the watermark bits stream. The new message digest MD' can be
calculated from the extracted watermark bits and the same is compared with extracted MD for
authentication.
Section 2.1 depicts the insertion technique.The re-adjustment of first transformed component has
been described in section 2.2 and the algorithm for extraction is given in section 2.3.
2.1 Insertion
Each 2 x 2 sub-matrices of the carrier image is converted into transformed components by
applying Binomial transform (BT) in a channel wise manner. One/two/three watermark bits are
embedded in second/third/fourth transformed components. An adjustment has been used to
reduce quality degradation by adjusting the embedded components without affecting the
fabricated bits. The inverse Binomial transform (IBT) is used to convert back the adjusted
components into pixel components.
Algorithm:
Input: The 128 bits message digest MD, derived from the authenticating watermark, the
carrier/cover image (I) and an authenticating watermark (message/image).
Output: The watermarked image (I′) in spatial domain.
Methods: The watermark (along with a message digest) is inserted into the carrier images by
converting the image from spatial domain to transform domain using Binomial transform (BT).
The detailed steps of embedding are as follows:
Steps:
1. Obtain 128 bits message digest MD from the authenticating watermark which may be
a textual message or an image.
2. Calculate the size of the authenticating watermark ((L = w + h), where w bits for
width and h bits for height). The authenticating watermark (W) bits size is:
Wsize=[B × {3 × (M × N)} – (MD + L)]
Where, the bits per byte (B) is 1.5; MD and L are the message digest and
dimension of the authenticating watermark respectively for the carrier image of size
M x N bytes. In our proposed technique, the MD and L consist of 128 and 32 bits.
3. Repeat step 3 until the message digests MD, the authenticating watermark size (Wsize)
and the authenticating watermark (W) is embedded.
a. The cover image (I) is partitioned into 2 x 2 non-overlapping blocks in row major
order. Each 2 x 2 sub-matrix (Sc) consists of four pixel components
corresponding to each channel.

b. The Binomial transform (BT) of equation (1) has been applied to convert each
pixel components into transformed components.
c. One/two/three bits from the message digest (MD), watermark size (Wsize) and the
watermark (W) are embedded on second/third/fourth transformed components
starting from the least significant bits position (LSB-0). Generally, if a
transformed component t consists of k bits and the number of bits inserted is r,
then the embedded transformed component t' becomes:
t' = {(t & (2k
- 2r
)) | d
The decimal value d, corresponding to r bit authenticating watermark, can be
expressed in binary form as, d = (ar-1ar-2ar-3……… ar-k)2, where for a positive
integer i, the value of (r-i) must be greater than or equal to 0, i.e., (r-i) ≥ 0.
d. A quality adjustment has been applied to get transformed component closest to the
original without hampering the fabricated bits. The adjustment has been done by
altering left most (k-r) bits and choosing the closest one of the original.
e. The adjusted components are converted back into the spatial domain using
inverse Binomial transform (IBT) of equation (2). During the inverse Binomial
transform (IBT), if any pixel component becomes negative or greater than 255,
then a transformed component re-adjustment methodology is applied on the first
transformed component discussed in section 2.2.
4. The successive bit embedding operations on each 2 x 2 sub-image block produces the
watermarked image (I′).
5. Stop.
2.2 Re-Adjustment
The Binomial Transform (BT) is used to convert the pixel components into transformed
components which can fabricate the authenticating watermark. The inverse Binomial transform
(IBT) may generate two serious problematic situations:
• The pixel component value may by negative (-ve).
• The pixel component value may be greater than the maximum value (i.e. 255).
The first situation can be handled by incrementing the first transformed component with one
while the second situation can be avoided by decrementing the first transformed component with
two, repeatedly.
2.3 Extraction
The watermarked image (I′) is received in spatial domain. During extraction, the watermarked
image has been taken as the input and the authenticating watermark size, content and message
digest MD are extracted. All extraction is done in transform domain from the transformed
components through reverse operation.

Algorithm:
Input: The watermarked image (I′).
Output: The watermark image (W) and the message digest.
Methods: The Binomial transform (BT) is used to extract the watermark (along with a
message digest) from the watermarked image by converting the image from spatial domain to
transform domain. Successive extracted bits forms the watermark data and generate a
message digest which can be used for authentication. The detailed steps of extraction are as
follows:
Steps:
1. Repeat step 1, until and unless the 128 bits message digest (MD), watermark
size(Wsize) and the authenticating watermark is extracted.
a. The watermarked image (I′) is partitioned into 2 x 2 non-overlapping blocks in
row major order. Each 2 x 2 sub-matrix (Sc) consists of four pixel components
corresponding to each channel.
b. The Binomial Transform (BT) of equation (1) has been applied to transform each
pixel components block (Sc) into transformed components.
c. One/two/three bits of the message digest (MD), watermark size (Wsize) and the
watermark(W) are extracted from second/third/fourth transformed components
starting from the least significant bits position (LSB-0). Each 8 (eight) bits
extraction ensures the construction of one alphabet/color component (R/G/B).
d. After extracting watermark bits, the transformed components are converted back
into the spatial domain using inverse Binomial transform (IBT) of equation (2).
Thus, the spatial domain sub-image block (S'c) is consists of four pixel
components while the bits are fabricated in transform domain.
2. Obtain 128 bits message digest MD′ from the extracted watermark.
3. Compare MD′ with extracted MD. If both are same then the image is authorized, else
unauthorized.
4. Stop.
3. RESULTS, COMPARISON AND ANALYSIS
Benchmark (PPM) images [4] of dimension 512 × 512 are taken to embed the gold coin (i.e. the
authenticating watermark image). Ten different color images (i-v) are taken. Images are labeled
as: (i) Lena, (ii) Baboon, (iii) Pepper, (iv) Earth, (v) Sailboat. On embedding the watermark
image i.e., the Gold-Coin image of (vi), the newly generated watermarked image produces a good
visual clarity.

(i) Lena (ii)
Baboon
(iii)
Pepper
(iv) Earth (v) Sailboat
(vi)
Gold
Coin
Figure 1. Different Cover images of dimension 512 × 512 along with the Authenticating Watermark Image
It is seen from table 1 that the watermarked images have a peak to signal noise ratio (PSNR) of
around 42 dB in average case whereas the average bits per byte (bpb) for a given carrier image is
1.5.
Table 1. Results of embedding of 147234 bytes of information in each image of dimension 512 x 512
Carrier Image Max. Payload
(byte)
PSNR IF BPB
Lena 147456 41.40 0.9997 1.5
Baboon 147456 42.05 0.9998 1.5
Pepper 147456 41.28 0.9996 1.5
Earth 147456 41.61 0.9997 1.5
Sailboat 147456 41.67 0.9998 1.5
AVG 147456 41.55 0.9997 1.5
On comparing the proposed technique with the SDHTIWCIA technique [1] and Varsaki et. al’s
[5] method, one can easily identify the enhancement of the PSNR at the same and higher payload.
Table 3. Comparison of BPB and PSNR for proposed technique over SDHTIWCIA [1] and Varsaki et.
Al’s [5] method
Carrier
Images
Varsaki et. al’s
Method [5]
SDHTIWCIA [1] Proposed
Technique
BPB
(bits per
byte)
PSNR
(dB)
BPB
(bits per
byte)
PSNR
(dB)
BPB
(bits per
byte)
PSNR
(dB)
Lena 0.25 39.70 1.5 37.95 1.5 41.40
Baboon 0.25 30.69 1.5 38.57 1.5 41.28
Sailboat 0.25 35.28 1.5 38.23 1.5 41.67
AVG 0.25 35.22 1.5 38.25 1.5 41.45
The histogram analysis ensures the minimal changes in the watermarked images. Fig-3 depicts
the histogram of ‘Lena’ image before and after embedding the ‘Gold Coin’ a separate channel
wise manner.

Histogram of Original Lena
Histogram of Watermarked Lena
Figure 3: Comparisons results of histogram between of original and watermarked image
It is seen from table 4 that the differences of mean, standard deviation and median between
original and watermarked images are very minimal.
Table 4. Comparison results of Mean, Median and Standard Deviation of original and watermarked images
Image Channel Mean Standard
Deviation
Median
Original
Lena
R 180.22 49.05 197
G 99.05 52.88 97
B 105.41 34.06 100
Watermarked
Lena
R 180.16 49.07 196
G 99.01 52.90 98
B 105.37 34.11 100
In this authentication system, the recipient operate the authentication process through matching of
the extracted message digest MD with the generated message digest MD' at the destination, where
MD' can be obtained from the extracted watermark image. If the extracted message digest MD
matches with the generated message digest MD', then the authentication process is said to be
successful, otherwise, it is said to be unsuccessful. That means any kind of attack on the
watermarked image is easily detectable.
The PSNR (Peak Signal to Noise Ratio) and NCC (Normalized Cross-correlation) values are
obtained from the watermarked images which are seen from table 5 by introducing attacks
namely ‘Median Filtering’, ‘Speckle Noise’ and ‘Salt & Pepper Noise’. It also ensures that the
qualities of attacked watermarked images are still well perceptible but the message digest ensures
that it is tampered.
Table 5. Comparison of PSNR and NCC values under different kinds of attacks for watermarked images
Images PSNR NCC PSNR NCC PSNR NCC PSNR NCC
Before
attack
After Median
Filtering attack ( 3
x 3 neighborhood)
After Speckle
Noise attack
(Variance =
0.001)
After Salt &
Pepper Noise
attack (Noise
Density=0.001)
Lena 41.40 0.9996 33.64 0.9960 34.19 0.9996 34.52 0.9995
Pepper 41.28 0.9996 31.84 0.9937 34.74 0.9997 34.00 0.9996
Sailboat 41.67 0.9998 28.52 0.9936 34.20 0.9996 34.35 0.9996

4. CONCLUSION
The proposed technique can be used to authenticate a color image by embedding watermark in
transformed components. Higher robustness is achieved by fabricating data in positive and
negative transformed components. The technique emphasizes on the improvement over
SDHTIWCIA [1] and Varsaki et. Al’s method [5].
ACKNOWLEDGEMENT
The authors express deep sense of gratuity towards the Dept of CSE University of Kalyani where
the computational resources are used for the work and the PURSE scheme of DST, Govt. of India
.
REFERENCES
[1] Mandal J.K, Ghosal, S.K “Separable Discrete Hartley Transform based Invisible Watermarking for
Color Image Authentication (SDHTIWCIA)”, Second International Conference on Advances in
Computing and Information Technology (ACITY-2012), July 13-15, Vol. 2, ISBN-978-3-642-31551-
0, pp. 767-776, Chennai, India, 2012.
[2] Borisov B. and Shkodrov V., Divergent Series in the Generalized Binomial Transform, Adv. Stud.
Cont. Math., 14 (1): 77-82, 2007.
[3] S. Falcon and A. Plaza, "Binomial transforms of the k-Fibonacci sequence", International Journal of
Nonlinear Sciences and Numerical Simulation 10(11-12): 1527-1538, 2009.
[4] Allan G. Weber, The USC-SIPI Image Database: Version 5, Original release: October 1997, Signal
and Image Processing Institute, University of Southern California, Department of Electrical
Engineering. http://sipi.usc.edu/database/ (accessed on 25th January, 2010).
[5] Varsaki et al, “On the use of the discrete Pascal transform in hiding data in images”, “Optics,
Photonics, and Digital Technologies for Multimedia Applications”, Proc. of SPIE Vol. 7723, 77230L
• © 2010 SPIE • CCC code: 0277-786X/10/$18 • doi: 10.1117/12.854220, 2010.

A fragile watermarking based on binomial transform in color images

More Related Content

What's hot (18)

Viewers also liked (20)

Similar to A fragile watermarking based on binomial transform in color images (20)

Recently uploaded (20)

A fragile watermarking based on binomial transform in color images