Professional Documents
Culture Documents
On
PRESENTED BY
1
deterring copyright theft of mapping data
watermarking has recently made its debut several newer techniques to protect
2
media possesses inherent advantages of difficult to remove without noticeably
portability, efficiency and accuracy of degrading the content and are a covert
information content in one hand, but on means in situations where cryptography
the other hand, this representation also fails to provide robustness.
puts a serious threat of easy, accurate and The content is watermarked by
illegal perfect copies of unlimited converting copyright information into
number. Unfortunately the currently random digital noise using a special
available formats for image, audio and algorithm that is perceptible only to the
video in digital form do not allow any content creator. Digital watermarks can be
type of copyright protection. A potential read only by using the appropriate reading
solution to this kind of problem is an software. These are resistant to filtering
electronic stamp or digital watermarking and stay with the content as long as
which is intended to complement Originally purposely degraded.
cryptographic process [1]. Digital watermarks don’t leave a
noticeable mark on the content and don’t
The technology: affect it’s appearance. These are
global network. It imposes extra degrading the content and are a covert
Digital watermarks don’t leave a read only by using the appropriate reading
noticeable mark on the content and don’t software. These are resistant to filtering
affect its appearance. These are and stay with the content as long as
3
While the later technique facilitates the information hidden in the higher
access of the encrypted data only for valid frequency components might be lost after
key holders but fails to track any quantization operation of lossy
reproduction or retransmission of data compression [3]. This motivates
after decryption. On the other hand, in researchers in recent times to realize the
digital watermarking, an identification importance of perceptual modeling of
code (symbol) is embedded permanently human visual system and the need to
inside a cover image which remains embed a signal in perceptually significant
within that cover invisibly even after regions of an image, especially if the
decryption process. This requirement of watermark is to survive lossy compression
watermarking technique, in general, needs [4]. In spatial domain block based
to possess the following characteristics: approach, this perceptually significant
(a) imperceptibility for hidden region is synonymous to low variance
information, (b) redundancy in blocks of the cover image.
distribution of the hidden information It is found in the literature that the
inside the cover image to satisfy robust watermarking systems proposed so
robustness in water mark extraction far can only withstand some of the
process even from truncated(cropped) possible external attacks but not all. While
image .and (c) one or more keys to spatial domain watermarking, in general,
achieve cryptographic security of hidden is easy to implement on computational
content [2]. Besides these general point of view but too fragile to withstand
properties, an ideal watermarking system large varieties of external attacks. On the
should also be resilient to insertion of other hand, frequency or Transformed
additional watermarks to retain the domain approach offers robust
rightful ownership. The perceptually watermarking but most cases
invisible data hiding needs insertion of implementationneed higher computational
watermark in higher spatial frequency of complexity. Moreover the transform
the cover image since human eye is less domain technique is global in nature
sensitive to this frequency component. (global within the block in block based
But in most of the natural images majority approach) and cannot restrict visual
of visual information are concentrated on degradation of the cover image. But in the
the lower end of the frequency band. So spatial domain scheme, degradation in
4
image quality due to watermarking could watermarked data/image. The generic
be controlled locally leaving the region of watermark recovery process needs the
interest unaffected. The present paper watermarked data, the secret key or public
describes a computationally efficient key and depending on the method, the
block based spatial domain watermarking original data and /or the original
technique for a two level watermark watermark as inputs while the output is
symbol. The selection of the required the recovered watermark W with some
block is based on variance of the block kind of confidence measure for the given
and watermark insertion exploits average watermark symbol or an indication about
brightness of the blocks. the presence of watermark in the cover
The Watermark recovery process document under inspection. Depending on
does not require either the the combination of inputs and outputs
cover/watermarked image or the three types namely private, semi private
watermark symbol only except the secret public watermarking system can be
image. The paper is organized as follows: defined [2].Private watermarking (also
section 2 describes the watermarking called non blind watermarking) requires at
principles. Section 3 describes insertion least the cover image and/or watermark
and extraction of watermark. Result is symbol and key (if used in embedding)
depicted in section 4 with conclusion in for the recovery of the hidden
section 5. information.Public watermarking (Blind
or oblivious watermarking) system
2. Watermarking principles: requires neither the cover image nor the
All watermarking methods share embedded watermark symbol but only the
the same building blocks [3]: an secret key/image during the detection of
5
Key and the watermark symbol but
without the cover image. Insertion of Watermark:
In the present work, a block based
. spatial domain algorithm is used to hide
copyright mark (invisible logo) in the
homogenous regions of the cover image
exploiting average brightness.
Step 1:
The cover image is partitioned into
non-overlapping square blocks of size
(8X8)
pixels. A block is denoted by the location
of its starting pixel (x, y). If the cover
image is of size (NXN), total (N/8XN/8)
number of such block is obtained for
watermark insertion. Next, all such blocks
are arranged in ascending order based on
their variance values. The variance (σ ²)
of a block of size (M X N) is denoted by
Figure2:Ge
m-1 n-1
neric water mark recovery scheme
σ ²=1/mn∑∑[ƒ(χ ,y)-µ ]² (1)
3 Insertion and Extraction of watermark x=0 y=0
The cover image I is a gray-level image of where
6
blocks and, of course, the smallness in present case is of length 256 and the
variance value depends on the spatially dispersed watermark data thus
characteristics of image to be obtained is denoted by L1.a J
watermarked. If the Watermark symbol is Step 3:
a (N X N) binary image, only N² From the two level image formed in
homogeneous blocks are sufficient to step 2, desired blocks Of the cover
insert one watermark pixel in each such image are selected and statistical average
homogenous block. A two level map of value of these blocks are used for
size (N/8XN/8) _is constructed based on watermark insertion. Let for one such
the location of homogenous blocks in the block this average value and its integer
cover image assigning each homogeneous part are denoted by A and A′ = A
block of the cover image by value ’1’ respectively. Now one pixel from L1
while all other blocks by value ’0’. This replaces a particular bit (preferably Least
two level map later modified as multi Significant Bit planes) in bit plane
level image, also called as secret image representation of A for each homogenous
(s), is used for extraction of watermark block. The selection of particular bit in bit
pixels. The formation of multilevel image plane representation may be determined
from two level maps is described in step based on the characteristics (busyness
3. /smoothness of regions) of the block. The
Step 2: bit plane selection is also governed by
In the proposed scheme, one watermark global characteristics of the cover image
pixel is inserted in each homogenous besides the local property of candidate
block. Before insertion, the binary block, such as mean gray value. For a
watermark is spatially dispersed using a block of low variance (homogenous zone)
chaotic system called” tours auto higher bit plane may be chosen provided
Orphism”. Basically, the tours auto that the mean gray level value of the
Orphism is a kind of image independent block is either less than T1 or greater than
permutation done by using pseudo T2, where T1 and T2 are certain pre-
random number of suitable length. This specified threshold values with T1 should
pseudo random number is generated preferably be close to ’0’ (minimum) and
using” Linear Feedback Shift Register”. T2 close to ’255’ (maximum). However,
The pseudorandom number in the the ’closeness’ of T1 and T2 to ’0’ and
7
’255’ respectively, is relative, and is done by estimating the tendency of
strongly image dependent. Users may possible change in mean gray value after
choose the value of T1 and T2 and also the attack like mean filtering. Larger size
the proper bit plane by checking the of spatial mask such as 7x 7 is used to
degradation in the image quality affected adjust suitably the gray values of all
by the insertion of the logo. pixels of the block. The use of spatial
A multilevel secret image is mask reduces visual distortion on and
constructed by inserting the value of bit average fifty percent times.
position selected for different Watermark Extraction:
homogeneous block located in the ’1’ The extraction of watermark requires
position of the secret image. This the secret image(s) and the key (k) used
positional information as gray value of the for spatial dispersion of the watermark
secret image helps to extract watermark image. The watermarked image under
pixel from the proper bit position of the inspection with or without external attacks
mean gray value of the block. Watermark is partitioned into non-overlapping block
insertion keeps all pixels values of each of size 8x8 pixels. Now from the secret
homogenous block unchanged, increased image, position of the homogenous blocks
or decreased by fixed value (based on the are selected and gray value of the secret
appropriate bit plane selection). image indicates the corresponding bit
Step 4: positioning mean gray values where
The choice of lower order MSB watermark pixel was inserted. Hence from
plane (say 3rd or higher from the bottom the secret image the mean gray value of
plane) may result in more robust the blocks of the watermarked
watermarking at the cost of greater visual image/distorted watermarked image is
distortion of the cover image. Further bit calculated and watermark pixel is
manipulation is done to minimize this extracted. The spatially dispersed
aberration and to counter the effect of watermark image thus obtained is once
smoothing that may cause possible loss of again permuted using the same key (k)
embedded information. The process (pseudo random number) and watermark
effectively changes those mean gray in original form is thus obtained. This
values of the blocks that have been used completes watermark extractionprocess.A
in watermark insertion. Implementation is quantitative estimation for the quality of
8
extracted watermark Image W(x,y) with of Watermarked image is 23.80dB and is
reference to the original watermark shown in Figure 5.
W(x,y) may be expressed as normalized Gaussian filtering:
cross correlation (NCC) Watermarked image
where_:olom! (PSNR=24.15dB) after two times
qpLpOn_ES\_]T__r_s_j__SG_]T__p L p Gaussian filtering with variance 1
O Q n__SG_UTV_rZ B (window size 9x9 ) isshown in Figure 6.
NCC= ∑x ∑y W(x, y) w′ (x, y)/ ∑x Figure 13 shows the extracted watermark
with NCC=0.88.
∑y [W(x, y)] ²
Median Filtering:
Gives maximum value of NCC as unity.
Watermarked image (PSNR=25.22
dB) obtained after five times median
Results:
filtering using a mask of size 3x3 is
Figure 3 shows Fishing boat
shown in Figure 7. Figure 14 shows
image used as cover image and Figure 4 is
extracted watermark image (NCC=0.94).
the watermarked image using logo/hidden
Image Rescaling:
symbol M as shown in Figure 11. Peak
The watermarked image was scaled
Signal to Noise Ratio (PSNR) of the
to one half of its original size and up
watermarked image to the original image
sampled to its original dimensions. Figure
is about 42.40 dB and hence quality
8 shows the modified image
degradations could hardly be perceived by
(PSNR=24.85 dB) with many details lost.
human eye. Robustness against different
Extracted watermark (with NCC=0.87) is
attacks is shown in table 1 and 2 for other
shown in Figure 15.
five test images such as Bear, New
York,Lena,Opera and Pills images shown
JPEG Compression:
in Figure 18,19,20,21 and 22 respectively
Figure 16 shows the extracted
[6,7].
watermark with NCC=0.958 from the
Mean Filtering:
watermarked image (PSNR=18.73 dB) as
Figure 12 shows extracted
shown in Figure 9 obtained after JPEG
watermark (NCC=0.80) from blurred
compression with compression ratio 45.0.
version of watermarked image (after mean
As compression ratio increases NCC
filtering) using 5x5 mask. PSNR value
value of the extracted watermark
9
decreases and the quality of the random fashion obtained by chaotic
watermark will also decrease accordingly. mixing.
10
Conclusion: Bibliography:
[1] R. Anderson. Information Hiding.
technique has been tested over large [3] Chiou-Ting Hsu and Ja-Ling Wu.
signal processing operations are found to [4] I.J. Cox, J. Kilian, T. Leighton and T.
scheme more resilient to other types of [5] S.Pereira, S.Voloshynoskiy and T.Pun.
11