NEURAL NETWORKS - A NEW DIMENSION IN DATA SECURITY

1. Introduction:

New possibilities of digital imaging and data hiding open wide prospects in modern imaging
science, content management and secure communications. However, despite the obvious
advantages of digital data hiding technologies and their current progress, these developments
carry inherent risks such as copyright violation, unauthorized prohibited usage and
distribution of digital media, secret communications and network security violations
.Although the issues of robustness, visibility and capacity of digital data hiding technologies
have received a lot of attention, their security aspect still remains an open and little studied
problem. The security requirement is closely related to the stochastic visibility and
unauthorized detection of hidden information and requires both new and careful study.
New information-theoretic methods for blind stochastic detection of hidden data should be
investigated. This aspect will have a great impact on robust digital watermarking,
steganography,integrity control and tamper proofing (possibly even without embedded
hidden data) and Internet/network security.

2.Related Research:
An introduction to steganography and its application to digital images are available from
[fig5]. Other, more robust, methods of hiding information in images include application of
the transform domain that takes advantage of algorithms and coefficients from processing the
image or its components to hide information. Transformations can be applied over the entire
image [8] to blocks through out the image [fig9, fig10], or other variations. Many of these
transformation techniques require use of the original, unmarked image to extract the
watermark. In [fig11] a number of papers propose techniques that do not require using the
original to extract the watermark [fig12]. A method that proposes a combination of these
techniques from LSB insertion to spread spectrum disbursement of data is described in [fig9].
A survey of transform domain techniques can be found in [fig13]. grabs. One common
drawback of virtually all current data embedding methods is the fact that the original image is
inevitably distorted due to data embedding itself. This distortion typically cannot be removed
completely due to quantization, bit-replacement, or truncation at the grayscales 0 and 255
[fig6]. Another unconventional approach in which we consider “piggy-backing” the color
information on the luminance component of an image for improved color image coding[fig1].
The new technique essentially transforms a given color image into the YIQ color space where
the chrominance information is subsampled and embedded in the wavelet domain of the
luminance component
3.The New Approach:

3.1 Multiple Key Function Method:

In this paper, we formulate a distinctive method for information hiding in images to beat the
existing detection techniques. Let us take the example of hiding an secret text called
the cipher text in an image called the cover image.

In the effort to keep the structural characteristics of the cover image, secure, we make
minimal introduction of message bits into the cover image and yet manage to embed text
messages of substantial size. It is thus possible, by choosing from a set of key functions,
which requires introducing least number of message bits in the cover image and still
accommodate the entire cipher intended to be hidden. The key functions are used to generate
a set of byte-positions where the message is to be embedded. At various bit positions in the
cover image a match between the cipher bit and the original image bit may result, which
means that actually, no replacement of the original image bit occurs. The Bit Replacement
Count (BRC) for this key function is calculated, dynamically, as the replacement/non-
replacement of the image bits occur. Here,

BRC = Number of replacements in the original image bits introduced

Now, another key function from the key function set is chosen and the above manipulations
are performed. As before, the new value of BRC is calculated. Redundant application of the
above formulation will result in a set of BRCs. The set of BRCs can be compared to find out
a key function, which belongs to the key function set, which causes least number of original
image bit replacements. The result is, a sizable text embedded into a cover image, without
actually disturbing much of the image‘s original characteristics.Associated with the key
functions, is a unique identification number in the key functions lookup table. Both the sender
and the receiver are required to have a similar lookup table. The unique key identification
number is also embedded in the cover image at specific positions in the cover image, along
with the cipher. The embedded cipher can be recovered safely by the receiver by first
extracting the key identification number from the cover image, and then generate the byte-
positions to actually decipher the embedded cipher information. Because the number of bits
in the cover retouched is minimal, as resulted from using the suitable key function, the image
can escape the statistical analyses (such as CHI-SQUARE or Extended CHI-SQUARE tests),

3.1.1 Implementation and Testing of Multiple Key method:

This approach was implemented and tested in windows environment with the c++ compiler.
The results given above are with a modest, 11 different key function table for a cipher text of
20 bytes. The perfect permutation of the key function and the cover image can produce
astonishing results with the ratio of the BRC to the size of the cipher, approaching unity.

BRC/Size of the Cipher = 1 (approx.)

The results ought to be better if the key function table is larger and a suitable database of
cover image is maintained. In such cases the probability of BRC would consistently reach
unity irrespective of the size of the cipher thus escaping all existing detection techniques.

fig 3
3.1.2Enhancements on this technique:

The complexity of the first approach is very high when the number of functions in the
function domain increases. The complexity can
be reduced by using a generator which dynamic all
generates functions in the neighborhood of the function that gives better BRC values
from a domain of just 50 functions.

3.2 Newton’s Forward Difference Method:

In this approach, Newton’s forward difference operator is introduced. The definition for that operator is

F(x) = F(x+h) - F(x), where h=interval of x values

In the method of Newton Forward Difference, by taking.

X :The serial number of cipher bits (1,2,3…n) [ always h=1 ].

F(x): The bit position of image where the respective cipher matches the original Image Bit

Using the above functional table, the forward difference table is constructed. From the table
thus obtained we get the polynomial function that can generate the functional table. This
polynomial function is then sent to the receiver along with the image. The receiver can
retrieve the polynomial function which is placed in some fixed positions in the image. He can
then generate the functional table thus knowing the bit positions where the cipher
information is present.
 3.2.1 Advantages of this method:
Since the method uses the original bits of the image to hide the cipher, the conventional
embedding is being ruled out and so probability of BRC is consistently equal to 1 for any
cipher size and any cover image. Since we are not going to choose all the bits of the image in
sequence, the size of the cipher that can be hidden is extremely large almost equal to half of
the size of the image. Since we are not going to embed, the statistical properties of the image
left unchanged. So the image escapes all the statistical detection techniques.

3.2.2 Implementation and testing:

This new approach was implemented and tested in windows environment with the c++
compiler. The forward difference table was constructed for a cipher length of 10 bytes., that
is, 80 bits and the functional table obtained is as given below.

fig 4

The above functional table yields the polynomial function 2x^5+2x^4-3x^2+2 after
constructing the forward difference table. The technique was tested for five cases by hiding
different ciphers in various images and the results thus obtained are shown in the graph.

Fig 5

Thus this approach proves to be more efficient than any other method because nowhere we
are going to replace or embed any data into the image. Any detention technique (CHI-
SQUARE method) fails to detect any suspection over the image.
3.2.3.Enhancements on this technique:
The degree of the polynomial thus obtained is very high .To reduce the degree of the
resulting Polynomial function, we use a selector to which some intelligence is incorporated.
We use the concepts of neural networks to achieve this task . We try to test and train the
selector so that it
intelligent selector such that the image bit positions such that the higher order differences in
theforward difference table vanishes to give a polynomial of lesser degree.

Fig 6

As shown above, the same technique is applied recursively on the resultant polynomial
functions Until the resultant polynomial function is of lesser degree. The polynomial table
resulting because
The enhancements is shown below. This table corresponds to the polynomial 3x-2.

fig 7
3.3.Comparison between the exitsting techniques and our method:

fig 8

4.Conclusion:

With the arrival of the digital society in the 21st century, the potential of information hiding
and watermarking techniques for multimedia has drawn increasingly intensive attention from
the research communities and commercial sectors all over the world during the last decade.
Recently, many researchers are investigating image data hiding techniques in various aspects
in addition to imperceptibility and robustness. These aspects include information theoretic,
stochastic, and security issues. This research trend will advance the information hiding
technologies and provide opportunity for practical systems. This paper proposes two new
evolutionary techniques in the field of data hiding. The first approach aims at minimal
embedding by increasing the number of functions. This approach provides facilities for using
65536 functions as we have used a 16 bit key.The second approach aims at avoiding
embedding thus satisfying theoretic, stochastic, and security issues.