Image Processing Reference
In-Depth Information
location in the same image. Even if this kind of forgery is easy to carry out, success-
ful detection is still a problem of great interest.
Thanks to its property that very simple rules can result in very complex be-
haviour, cellular automata (CA) are commonly used for various image processing
tasks. The basic idea for using a CA for CMF detection is based on an assumption
that similar areas on the image should produce similar rules. The main approach is
one variation of block-based methods, where the image is divided into overlapping
blocks and each block's properties are analysed. Thereby, CA are applied on every
overlapping block of forged image with the aim to generate a set of rules. This pro-
cess can be present as a selection of a subset of rules that describe the texture of
block from all possible rules.
Application of a CA on a greyscale image leads to a large number of possible
rules and an even larger number of possible subsets of those rules. Reduction of
number of rules can be accomplished by a proper binary representation of image,
resulting in only two possible values of cells states (as opposed to 256 in case when
a greyscale image is used). Thresholding of a greyscale image by global threshold
leads to binary image where much information about texture is lost, and usage of
binary planes is highly noise sensitive. In order to solve these issues, a new rep-
resentation of the image based on local binary pattern (LBP) is introduced. LBP
defines binary values of neighbourhood pixels based on a difference between cen-
tral and neighbourhood pixels. Consequently, LBP preserves local information but
also keeps enough global images' information.
Detection of copy-move forgery is accomplished using simple 1D CA where the
neighbourhood for every pixel is defined as a group of pixels from the row above
the pixel under consideration. In our experiments, a neighbourhood size of 7 pix-
els showed best results according to computational time and detection accuracy.
Thresholding to binary values is based on the mean value of all pixels in the neigh-
bourhood and the pixel under consideration. The method is very reliable in different
testing conditions, but there are some cases when detection is difficult, such as in
the presence of homogeneous areas with small difference between pixels values. In
those situations many blocks are false detected as forged. The problem is caused by
the presence of similar values of pixels, leading to the same binary representation
and the same set of rules for different blocks.
The presented method shows robustness to blurring of the image with averaging
filters, but adding noise strongly affects the detection results. However, coping with
noise is possible by a simple pre-processing using an averaging filter to smooth the
image prior to applying the CA.
References
1. Amerini, I., Ballan, L., Caldelli, R., Bimbo, A.D., Serra, G.: A SIFT-based forensic
method for copy-move attack detection and transformation recovery. IEEE Transactions
on Information Forensics and Security 6(3), 1099-1110 (2011)
2. Bashar, M., Noda, K., Ohnishi, N., Mori, K.: Exploring duplicated regions in natural
images. IEEE Transactions on Image Processing (2010) (accepted for publication)
Search WWH ::
Custom Search