Documente Academic
Documente Profesional
Documente Cultură
dt e t w t x
ft j
2 *
] ) ( ) ( [ ------------ (2.1)
Where the length of the window is (t-) in time such that we can shift the window by
changing value of t, and by varying the value we get different frequency response of the
signal segments.
18
The Heisenberg uncertainty principle explains the problem with STFT. This principle
states that one cannot know the exact time-frequency representation of a signal, i.e., one cannot
know what spectral components exist at what instances of times. What one can know are the
time intervals in which certain band of frequencies exists and is called resolution problem.
This problem has to do with the width of the window function that is used, known as the support
of the window. If the window function is narrow, then it is known as compactly supported. The
narrower we make the window, the better the time resolution, and better the assumption of the
signal to be stationary, but poorer the frequency resolution:
Narrow window ===> good time resolution, poor frequency resolution
Wide window ===>good frequency resolution, poor time resolution
The wavelet transform (WT) has been developed as an alternate approach to STFT to
overcome the resolution problem. The wavelet analysis is done such that the signal is multiplied
with the wavelet function, similar to the window function in the STFT, and the transform is
computed separately for different segments of the time-domain signal at different frequencies.
This approach is called Multi resolution Analysis (MRA) [4], as it analyzes the signal at
different frequencies giving different resolutions.
MRA is designed to give good time resolution and poor frequency resolution at high
frequencies and good frequency resolution and poor time resolution at low frequencies. This
approach is good especially when the signal has high frequency components for short durations
and low frequency components for long durations, e.g., images and video frames.
The wavelet transform involves projecting a signal onto a complete set of translated
and dilated versions of a mother wavelet +(t). The strict definition of a mother wavelet will be
dealt with later so that the form of the wavelet transform can be examined first. For now,
assume the loose requirement that +(t) has compact temporal and spectral support (limited by
the uncertainty principle of course), upon which set of basis functions can be defined.
The basis set of wavelets is generated from the mother or basic wavelet is defined as:
+a,b(t) = |
.
|
\
|
a
b t
a
1
; a, b e 9 and a>0 ------------ (2.2)
19
The variable a (inverse of frequency) reflects the scale (width) of a particular basis
function such that its large value gives low frequencies and small value gives high frequencies.
The variable b specifies its translation along x-axis in time. The term 1/ a is used for
normalization.
3.2. WAVELET TRANSFORM
Whether we like it or not we are living in a world of signals. Nature is talking to us with
signals: light, sounds Men are talking to each other with signals: music, TV, phones, etc.
The human body is equipped to survive in this world of signals with sensors such as
eyes and ears, which are able to receive and process these signals. Consider, for instance, our
ears: they can discriminate the volume and tone of a voice. Most of the information our ears
process from a signal is in the frequency content of the signal.
Scientists have developed mathematical methods to imitate the processing performed
by our body and extract the frequency information contained in a signal. These mathematical
algorithms are called transforms and the most popular among them is the Fourier Transform.
The second method to analyze non-stationary signals is to first filter different frequency
bands, cut these bands into slices in time, and then analyze them.
The wavelet transform uses this approach. The wavelet transform or wavelet analysis
is probably the most recent solution to overcome the shortcomings of the Fourier transform. In
wavelet analysis the use of a fully scalable modulated window solves the signal-cutting
problem. The window is shifted along the signal and for every position the spectrum is
calculated. Then this process is repeated many times with a slightly shorter (or longer) window
for every new cycle.
In the end the result is a collection of time-frequency representations of the signal, all
with different resolutions. Because of this collection of representations, we can speak of a
multi resolution analysis. In the case of wavelets, we normally do not speak about time-
frequency
3.3 Discrete Wavelet Transform
The discrete wavelet transform (DWT) was developed to apply the wavelet transform
to the digital world. Filter banks are used to approximate the behavior of the continuous wavelet
transform. The signal is decomposed with a high-pass filter and a low-pass filter. The
20
coefficients of these filters are computed using mathematical analysis and made available to
you. See Appendix B for more information about these computations.
Fig 3.1 Discrete Wave Transform
Where,
LPd: Low Pass Decomposition Filter
HPd: High Pass Decomposition Filter
LPr: Low Pass Reconstruction Filter
HPr: High Pass Reconstruction Filter
The wavelet literature presents the filter coefficients to you in tables. An example is
the Daubechies filters for wavelets. These filters depend on a parameter p called the
vanishing moment.
Table 3.1 Daubechies Wavelet Coefficient
The hp[n] coefficients are used as the low-pass reconstruction filter (LPr).
The coefficients for the filters HPd, LPd and HPr are computed from the h[n] coefficients as
follows:
High-pass decomposition filter (HPd) coefficients
g[n] =(1)n h[Ln] (L: length of the filter)
Low-pass reconstruction filter (LPr) coefficients
h[n] =h[Ln] (L: length of the filter)
21
High-pass reconstruction filter (HPr) coefficients
g[n] =g[Ln] (L: length of the filter)
The Daubechies filters for Wavelets are provided in the C55x IMGLIB for 2 p 10.
Since there are several sets of filters, we may ask ourselves what are the advantages and
disadvantages to using one set or another.
First we need to understand that we will have perfect reconstruction no matter what the
filter length is. However, longer filters provide smoother, smaller intermediate results. Thus, if
intermediate processing is required, we are less likely to lose information due to necessary
threshold or saturation. However, longer filters obviously involve more processing.
3.4 Wavelets and Perfect Reconstruction Filter Banks
Filter banks decompose the signal into high- and low-frequency components. The low-
frequency component usually contains most of the frequency of the signal. This is called the
approximation. The high-frequency component contains the details of the signal.
Wavelet decomposition can be implemented using a two-channel filter bank. Two-
channel filter banks are discussed in this section briefly. The main idea is that perfect
reconstruction filter banks implement series expansions of discrete-time signals.
Fig 3.2 A two level Wavelet Decomposition
Fig 3.3 A two level Wavelet Reconstruction
22
The input and the reconstruction are identical; this is called perfect reconstruction. Two
popular decomposition structures are pyramid and wavelet packet. The first one decomposes
only the approximation (low-frequency component) part while the second one decomposes
both the approximation and the detail (high-frequency component).
Fig 3.4 Pyramid Packet
Fig3.5. Wavelet Packet Decomposition
The C55x IMGLIB provides the following functions for one dimension pyramid and
packet decomposition and reconstruction. Complete information about these functions can be
found in the C55x IMGLIB.
1-D discrete wavelet transform
void IMG_wave_decom_one_dim(short *in_data, short *wksp, int *wavename, int length,int
level);
23
1-D inverse discrete wavelet transform
void IMG_wave_recon_one_dim(short *in_data, short *wksp, int *wavename, int length,int
level);
1-D discrete wavelet package transform
void IMG_wavep_decom_one_dim(short *in_data, short *wksp, int *wavename, int
length,int level);
1-D inverse discrete wavelet package transform
void IMG_wavep_recon_one_dim(short *in_data, short *wksp, int *wavename, int length,int
level);
3.5 Wavelets Image Processing
Wavelets have found a large variety of applications in the image processing field. The
J PEG 2000 standard uses wavelets for image compression. Other image processing
applications such as noise reduction, edge detection, and finger print analysis have also been
investigated in the literature.
3.6 Wavelet Decomposition of Images
In wavelet decomposing of an image, the decomposition is done row by row and then
column by column. For instance, here is the procedure for an N x M image. You filter each
row and then down-sample to obtain two N x (M/2) images. Then filter each column and
subsample the filter output to obtain four (N/2) x (M/2) images.
Of the four subimages obtained as seen in Figure 12, the one obtained by low-pass
filtering the rows and columns is referred to as the LL image.
The one obtained by low-pass filtering the rows and high-pass filtering the columns is
referred to as the LH images. The one obtained by high-pass filtering the rows and low-pass
filtering the columns is called the HL image. The subimage obtained by high-pass filtering the
rows and columns is referred to as the HH image. Each of the subimages obtained in this
fashion can then be filtered and subsampled to obtain four more subimages. This process can
be continued until the desired subband structure is obtained.
24
Fig 3.6 Original Image One-Level 2-D Decomposition
Three of the most popular ways to decompose an image are: pyramid, spacl, and wavelet
packet, as shown in Fig 3.7.
Fig 3.7 Three Popular Wavelet Decomposition Structures on Image
(a) Pyramid (b) Spacl, (c) Wavelet Packet
In the structure of pyramid decomposition, only the LL subimage is decomposed after each
decomposition into four more subimages.
In the structure of wavelet packet decomposition, each subimage (LL, LH,HL, HH) is
decomposed after each decomposition.
In the structure of spacl, after the first level of decomposition, each subimage is decomposed
into smaller subimages, and then only the LL subimage is decomposed.
In the part I development stage, the J PEG 2000 standard supports the pyramid
decomposition structure. In the future all three structures will be supported.
For two dimensions, the C55x IMGLIB provides functions for pyramid and packet
decomposition and reconstruction. Complete information about these functions can be found
in the C55x IMGLIB.
2-D discrete wavelet transform
void IMG_wave_decom_two_dim(short **image, short * wksp, int width, int height, int
*wavename, int level);
2-D inverse discrete wavelet transform
25
void IMG_wave_recon_two_dim(short **image, short * wksp, int width, int height, int
*wavename, int level);
2-D discrete wavelet package transform
void IMG_wavep_decom_two_dim(short **image, short * wksp, int width, int height, int
*wavename, int level);
2-D inverse discrete wavelet package transform
void IMG_wavep_recon_two_dim(short **image, short * wksp, int width, int height, int
*wavename, int level);
3.7 1-D Continuous wavelet transform
The 1-D continuous wavelet transform is given by:
Wf (a, b) =
}
dt t t x
b a
) ( ) (
,
------------ (2.3)
The inverse 1-D wavelet transform is given by:
x (t) =
} }
0
2
,
) ( ) , (
1
a
da
db t b a W
C
b a f
------------ (2.4)
Where C =
d
2
)
<
( ) + is the Fourier transform of the mother wavelet +(t). C is required to be finite,
which leads to one of the required properties of a mother wavelet. Since C must be finite, then
0 ) 0 ( = + to avoid a singularity in the integral, and thus the ) (t + must have zero mean. This
condition can be stated as
}
dt t) (
=0 and known as the admissibility condition.
26
1-D Discrete wavelet transform
The discrete wavelets transform (DWT), which transforms a discrete time signal to a
discrete wavelet representation. The first step is to discretize the wavelet parameters, which
reduce the previously continuous basis set of wavelets to a discrete and orthogonal /
orthonormal set of basis wavelets.
m,n(t) =2
m/2
(2
m
t n) ; m, n e Z such that - <m, n < -------- (2.5)
The 1-D DWT is given as the inner product of the signal x(t) being transformed with
each of the discrete basis functions.
Wm,n =<x(t), m,n(t) > ; m, n eZ ------------ (2.6)
The 1-D inverse DWT is given as:
x (t) =
m n
n m n m
t W ) (
, ,
; m, n eZ ------------- (2.7)
One Dimension Wavelet Applications
The 1D_Demo.c file presents applications of the one-dimension wavelet. A 128-point
sine wave is used as input for all these applications.
1-D Perfect Decomposition and Reconstruction Example
The third application shows a three-level pyramid decomposition and reconstruction
of the input signal:
// Perfect Reconstruction of Pyramid, Level 3
//==================================================
for( i =0; i <LENGTH; i++)
signal[i] =backup[i];
IMG_wave_decom_one_dim( signal, temp_wksp, db4, LENGTH, 3 );
27
IMG_wave_recon_one_dim( signal, temp_wksp, db4, LENGTH, 3 );
for( i =0; i <LENGTH; i++)
noise[i] =signal[i] backup[i];
//
3.9 2-D wavelet transform
The 1-D DWT can be extended to 2-D transform using separable wavelet filters. With
separable filters, applying a 1-D transform to all the rows of the input and then repeating on all
of the columns can compute the 2-D transform. When one-level 2-D DWT is applied to an
image, four transform coefficient sets are created. As depicted in Figure 3.8 (c), the four sets
are LL, HL, LH, and HH, where the first letter corresponds to applying either a low pass or
high pass filter to the rows, and the second letter refers to the filter applied to the columns.
Figure 3.8. Block Diagram of DWT (a)Original Image (b) Output image after the 1-D applied
on Row input (c) Output image after the second 1-D applied on row input
Figure 3.9. DWT for Lena image (a)Original Image (b) Output image after the 1-D
applied on column input (c) Output image after the second 1-D applied on row input
28
The Two-Dimensional DWT (2D-DWT) converts images from spatial domain to
frequency domain. At each level of the wavelet decomposition, each column of an image is
first transformed using a 1D vertical analysis filter-bank. The same filter-bank is then applied
horizontally to each row of the filtered and subsampled data. One-level of wavelet
decomposition produces four filtered and subsampled images, referred to as subbands. The
upper and lower areas of Fig. 3.9(b), respectively, represent the low pass and high pass
coefficients after vertical 1D-DWT and sub sampling. The result of the horizontal 1D-DWT
and sub sampling to form a 2D-DWT output image is shown in Fig.3.9(c).
We can use multiple levels of wavelet transforms to concentrate data energy in the
lowest sampled bands. Specifically, the LL subband in fig 3.8(c) can be transformed again to
form LL2, HL2, LH2, and HH2 subbands, producing a two-level wavelet transform. An (R-1)
level wavelet decomposition is associated with R resolution levels numbered from 0 to (R-1),
with 0 and (R-1) corresponding to the coarsest and finest resolutions.
The straight forward convolution implementation of 1D-DWT requires a large amount
of memory and large computation complexity. An alternative implementation of the 1D-DWT,
known as the lifting scheme, provides significant reduction in the memory and the computation
complexity. Lifting also allows in-place computation of the wavelet coefficients. Nevertheless,
the lifting approach computes the same coefficients as the direct filter-bank convolution.
In this application, the image is one-level decomposed and reconstructed. You notice
no difference between the original picture and the reconstructed picture.
3.10. LIFTING USING HARR
The lifting scheme is a useful way of looking at discrete wavelet transform. It is easy
to understand, since it performs all operations in the time domain, rather than in the frequency
domain, and has other advantages as well. This section illustrates the lifting approach using the
Haar Transform [6].
The Haar transform is based on the calculations of the averages (approximation co-
efficient) and differences (detail co-efficient). Given two adjacent pixels a and b, the principle
is to calculate the average
2
) ( b a
s
+
= and the difference b a d = . If a and b are similar, s will
29
be similar to both and d will be small, i.e., require few bits to represent. This transform is
reversible, since
2
d
s a = and
2
d
s b + = and it can be written using matrix notation as
\
|
=
2 / 1
2 / 1
) , ( ) , ( b a d s
|
|
.
|
1
1
=(a,b)A,
\
|
=
2 / 1
1
) , ( ) , ( d s b a
|
|
.
|
2 / 1
1
=
1
) , (
A d s
Consider a row of
n
2 pixels values
l n
S
,
for
n
l 2 0 < s . There are
1
2
n
pairs of pixels
2
1 2 , 2 ,
2 ,....., 4 , 2 , 0 ,
+
=
n
l n l n
forl S S . Each pair is transformed into an average
2 / ) (
1 2 , 2 , 1 , 1 +
+ =
l n l n n
S S S and the difference
l n l n l n
S S d
2 , 1 2 , , 1
=
+
. The result is a set
1
1
2
n
n
of S
averages and a set
1
1
2
n
n
of d differences.
3.11 Introduction of LSB
Data hiding is a method of hiding secret messages into a cover-media such that an
unintended observer will not be aware of the existence of the hidden messages. In this paper,
8-bit grayscale images are selected as the covermedia. These images are called cover-images.
Cover-images with the secret messages embedded in them are called stego-images. For data
hiding methods, the image quality refers to the quality of the stego-images.
In the literature, many techniques about data hiding have been proposed. One of the
common techniques is based on manipulating the least-signi7cant-bit (LSB) planes by
directly replacing the LSBs of the cover-image with the message bits. LSB methods typically
achieve high capacity. Wang et al. proposed to embed secret messages in the moderately
significant bit of the cover-image. A genetic algorithm is developed to 7nd an optimal
substitution matrix for the embedding of the secret messages.
They also proposed to use a local pixel adjustment process (LPAP) to improve the
image quality of the stego-image. Unfortunately, since the local pixel adjustment process
only considers the last three least significant bits and the fourth bit but not on all bits, the
local pixel adjustment process is obviously
30
not optimal. The weakness of the local pixel adjustment process is pointed out in Ref. As the
local pixel adjustment process modifies the LSBs, the technique cannot be applied to data
hiding schemes based on simple LSB substitution. Recently, Wang et al. further proposed a
data hiding scheme by optimal LSB substitution and genetic algorithm. Using the proposed
algorithm, the worst mean-square-error (WMSE) between the cover-image and the stego-
image is shown to be 1/2 of that obtained by the simple LSB substitution method.
In this paper, a data hiding scheme by simple LSB substitution with an optimal pixel
adjustment process (OPAP) is proposed. The basic concept of the OPAP is based on the
technique proposed. The operations of the OPAP are generalized. The WMSE between the
cover-image and the stego-image is derived. It is shown that the WMSE obtained by the OPAP
could be less than of that obtained by the simple LSB substitution method. Experimental
results
demonstrate that enhanced image quality can be obtained with low extra computational
complexity. The results obtained also show better performance than the optimal substitution
method described.
3.12 Data hiding by simple LSB substitution
In this section, the general operations of data hiding by simple LSB substitution
method are described.
Let C be the original 8-bit grayscale cover-image of Mc Nc pixels represented as
M be the n-bit secret message represented as
Suppose that the n-bit secret message M is to be embedded into the k-rightmost LSBs of the
cover-image C. Firstly, the secret message M is rearranged to form a conceptually k-bit
virtual image M_ represented as
Where The mapping between the n-bit secret message.
31
And the embedded message can be defined as follows:
Secondly, a subset of n_ pixels {xl1; xl2 ,.. xln_ }is chosen from the cover-image C in a
predefined sequence. The embedding process is completed by replacing the k LSBs of xli by
mi . Mathematically, the pixel value xli of the chosen pixel for storing the k-bit message mi
is modified to form the stego-pixel xli as follows:
In the extraction process, given the stego-image S, the embedded messages can be readily
extracted without referring to the original cover-image. Using the same sequence as in the
embedding process, the set of pixels {xl1, xl2,. Xln}storing the secret message bits are
selected from the stego-image. The k LSBs of the selected pixels are extracted and lined up to
reconstruct the secret message bits. Mathematically, the embedded message bits mi can be
recovered by
Suppose that all the pixels in the cover-image are used for the embedding of secret message
by the simple LSB substitution method. Theoretically, in the worst case, the
PSNR of the obtained stego-image can be computed by
32
Table 1 tabulates the worst PSNR for some k =15. It could be seen that the image quality of
the stego-image is degraded drastically when k>4.
3.13 Optimal pixel adjustment process
In this section, an optimal pixel adjustment process (OPAP) is proposed to enhance
the image quality of the stego-image obtained by the simple LSB substitution method. The
basic concept of the OPAP is based on the technique proposed.
Let pi , pi and pi be the corresponding pixel values of the ith pixel in the cover-image C,
the stego-image C obtained by the simple LSB substitution method and the refined stego-
image obtained after the OPAP. Let
Be the embedding error between pi and pi . According to the embedding process of the
simple LSB substitution method described in Section 2, pi is o
The value of can be further segmented into three intervals, such that
Based on the three intervals, the OPAP, which modifi7es pi to form the stego-pixel pi, can
be described as follows:
33
Let be the embedding error between pi and pi. can be computed as
follows:
From the above 7ve cases, it can be seen that the absolute value of may fall into the range
only when and ( case
5);while for other possible values of falls into the range , because
is obtained by the direct replacement of the k lab of Pi with the message bits
34
are equivalent to Pi<2^k and Pi >256-2^k, respectively in
general for gray scale natural images, when k<4, the no of pixel with pixel value smaller than
2^k or greater than 256-2^k, is insignificant. As a result it could be estimated that the absolute
embedded error between pixel in the cover image and in the stego image obtained after the
proposed OPAP is limited to
Let WMSE and WMSE be the worst case mean-squareerror between the stego-image
and the cover-image obtained by the simple LSB substitution method and the proposed method
with OPAP, respectively. WMSE* can be derived by
Reveals that WMSE*<WMSE, for k>2; and WMSE* WMSE when k=4. This
result also shows that the WMSE* obtained by the OPAP is better than that obtained by the
optimal substitution method proposed in which WMSE* = WMSE.
Moreover, the optimal pixel adjustment process only requires a checking of the
embedding error between the original cover-image and the stego-image obtained by the
simple LSB substitution method to form the final stego-image.
The extra computational cost is very small compared with Wangs method which requires
huge computation for the genetic algorithm to 7nd an optimal substitution matrix.
35
CHAPTER 4
RESULT AND DI SCUSSI ON
4.1 Advantages
1. High hiding capacity & Imperceptibility
2. Less Embedding Error
3. Here we avoided overflow & underflow error while embedding
4. High Robustness
4.2 Applications
1. Teleconference or video phone
- Very low delay (1/10 second is a standard)
2. Live Broad cast Video
- Modest delay is tolerable(second is normal)
- Error tolerance is needed
3. Video-in-a-can (DVD, Video on Demand)
- Random access to compressed data is desired
- Encoding can take a lot time
4. Decoding must always be at least the frame rate.
4.3 Result
Fig 4.1 Result Data hiding in Video
In general the ratio of PSNR should above 40 and we are obataining PSNR ratio above 60 so
PSNR is better
36
CHAPTER 5
CONCLUSI ON
In this paper we proposed a novel data hiding scheme that hides data into the integer
wavelet coefficients of an image. The system combines an adaptive data hiding technique and
the optimum pixel adjustment algorithm to increase the hiding capacity of the system
compared to other systems. The proposed system embeds secret data in a random order using
a secret key only known to both sender and receiver.
It is an adaptive system which embeds different number of bits in each wavelet
coefficient according to a hiding capacity function in order to maximize the hiding capacity
without sacrificing the visual quality of resulting stego image. The proposed system also
minimizes the difference between original coefficients values and modified values by using
the optimum pixel adjustment algorithm.
The proposed scheme was classified into three cases of hiding capacity according to
different applications required by the user. Each case has different visual quality of the stego-
image. Any data type can be used as the secret message since our experiments was made on a
binary stream of data. There was no error in the recovered message (perfect recovery) at any
hiding rate. From the experiments and the obtained results the proposed system proved to
achieve high hiding capacity up to 48% of the cover image size with reasonable image
quality and high security because of using random insertion of the secret message. On the
other hand the system suffers from low robustness against various attacks such as histogram
equalization and J PEG compression.
The proposed system can be further developed to increase its robustness by using
some sort of error correction code which increases the probability of retrieving the message
after attacks, also investigating methods to increase visual quality of the stego-image (PSNR)
with the obtained hiding capacity.
Hope to add support files to hide all file formats. This allows for much broader
spectrum of uses that one would able to encode .mp4,mov, 3gp,.
37
REFERENCES
1. G. J . Simmons, "The prisoners' problern and the subliminal channel," in Proceedings
of Crypto' 83, pp. 51-67, 1984.
2. N. Wu and M. Hwang. "Data Hiding: Current Status and Key Issues," International
J ournal of Network Security, Vol.4, No.1, pp. 1-9, J an. 2007.
3. W. Chen, "A Comparative Study of Information Hiding Schernes Using Amplitude,
Frequency and Phase Embedding," PhD Thesis, National Cheng Kung University,
Tainan, Taiwan, May 2003.
4. C. Chan and L. M. Cheng, "Hiding data in images by simple LSB substitution,
Pattern Recognition, pp. 469-474, Mar. 2004.
5. Changa, C. Changa, P. S. Huangb, and T. Tua, "A Novel bnage Steganographic
Method Using Tri-way Pixel-Value Differencing," J ournal of Multimedia, Vol. 3,
No.2, June 2008.
6. H. H. Zayed, "A High-Hiding Capacity Technique for Hiding Data in images Based
on K- Bit LSB Substitution," The 30th International Conference on Artificial
Intelligence Applications (ICAIA 2005) Cairo, Feb. 2005.
7. S. Katzenbeisser and F.A.P. Petitcolas, Information Hiding Techniques for
Steganography and Digital Watermarking, Artech House, Boston, 2000.
8. H.K. Pan, Y.Y. Chen, and Y.C. Tseng, A Secure Data Hid-ing Scheme for Two-
Color Images, Proc. Fifth IEEE Symp. Computers and Comm., IEEE Press,
Piscataway, N.J ., 2000
9. S. Lee, C.D. Yoo and T. Kalker, "Reversible image watermarking based on integer-
to-integer wavelet transform," IEEE Transactions on Information Forensics and
Security, Vol. 2, No.3, Sep. 2007, pp. 321-330.
10. M. K. Ramani, Dr. E. V. Prasad and Dr. S. Varadarajan, "Steganography Using
BPCS to the Integer Wavelet Transformed bnage", UCSNS International J ournal
of Computer Science and Network Security, VOL. 7 No.7, J uly 2007.
11. A. R. Calderbank, 1. Daubechies, W. Sweldens and B. Yeo., "Wavelet transforms
that map integers to integers". Applied and Computational Harmonic Analysis, vol.5,
noJ , pp.332-369, 1998.
12. G. Xuan, J . Zhu, Y. Q. Shi, Z. Ni, and W. Su., "Distortionless data hiding based on
integer wavelet transform," IEE Electronic Letters, 38(25): 1646--1648, Dec. 2002.