03-10-2012, 12:21 PM
A New Int2Int High Capacity Robust Steganography Method with LSB 1/3 and rounding method for embedding message
A New Int2Int High.pdf (Size: 244.24 KB / Downloads: 71)
Abstract:
In this paper a new way for hiding message in digital
images is proposed, that is robust against ALE and past other
steganalysis methods. It has high capacity and good perceptual
transparency and robustness. Proposed steganography method
uses 2D int2int wavelet transform and based on Ramani’s idea,
divides sub bands to 8×8 blocks and constructs bit planes of
each block, uses the cawagushi criteria for computing bit plane
complexity and then computes the capacity of block by
founding the first most significant bit that has complexity more
than threshold. If the capacity is 1, uses the LSB 1/3 and if it is
more than 1 uses rounding method for embedding message.
Capacity of block is saved in three least significant bits of first
value of block. After each embedding in each block, ALE
steganalysis feature vectors are computed and if they change
more than specified number TE, we ignore embedding in this
block, save the value 0 or “000” in first value of block and
replace the main numbers in block. Keep embedding from next
block according to pseudo random number generator stream.
At final, calculate inverse 2-D int2int wavelet transform and
obtains the stego image.
Introduction
Steganography is art and science of hiding message in
digital images. The most important aim of steganography
is that the existence of message in digital messages
shouldn’t be discoverable. There are different ways for
embedding message in digital Images. Against of this
Steganography methods Steganalysers are trying to
detect the existence of hidden message. In this paper a
new algorithm for embedding message in digital images
is introduced. This method works in integer Wavelete
transform domain [2]. Pixel domain steganography
methods have little complexity relate to transform
domain and more capacity but often.
Experimental result
For compromising our steganography system against
ALE steganalysis method we use ROC figures that
obtained from training and testing two layer neural
networks with linear first layer and nonlinear hidden
layer that has 20 neurons in hidden layer. Inputs are
1500, (10×1) feature vectors and 1500, (2×1) class
vectors. 70% of data for train, 15% for validation and
remained other 15% are used for testing. This neural
network is tested with Ramani and proposed
steganography method.
Result and Feature Works
As result show the proposed steganography has good
perceptual transparency and high capacity for embedding
and is robust against the ALE steganalysis system. For
future work we can test our algorithm on camera
database and NRCS data base of image and compare the
result. Also we can use another transform domain as
curvlet or ridgelet instead of 2-D int2int wavelet
transform. Also for increasing the capacity in state that
we embedding in one block change the ALE feature
vector more than TE we can reduce the capacity by one
and use the embedding algorithm again with new
capacity instead of ignoring this block of embedding.