We consider an embedding process to counterfoil the detection process using Independent Component Analysis (ICA) and present a method that demonstrates how ICA can be used to detect forged documents. The expansion of network systems and the lack of difficulty in reproducing digital media which does not lack quality have necessitated novel techniques to restrain the forbidden reproduction, forgery and dissemination of digital data [191], [192], [193] and [194]. Watermarking and cryptography are devices that assist the enforcing of copyright protection and authentication of digital audio, images and video. For the purposes of this investigation we focused on images only.
Robust digital watermarks used for copyright protection can be explained as undetectable information concealed in a digital media. This information should be detectable as long as the quality of the content is considered as being acceptable. Despite many algorithms, methods, techniques and fully functional systems being used to hide information, the main dilemma is that the bulk of these methods use a symmetric key [195].
Both symmetric watermarks and symmetric cryptography is similar: whereby the exact key is used to encode the information and to authenticate the watermark. The key becomes
Figure 5.15: Example of passport authentication system (GUI) and a decrypt of the texture code (top-left image) generated by diffusing the cipher (top-right image) with the passport holders details revealed in the decrypt as shown (bottom image). Note that this figure is grey scale images of a colour GUI and colour images.
the ‘weakest link’. In order to decipher or decode the hidden data it is necessary to know the secret key [195]. Once the secret key is known, the watermark can be not only decoded, but also easily estimated and removed from the content. In such a state of affairs, a decoder used for copy protection has to be either put into operation as a tamper-proof device, or located in a trusted third party [196]. These solutions are expensive and not necessarily plausible.
We used ICA to separate our 2D DataGlyph from the watermark itself. Of interest is the fact that the 2D DataGlyph plays the role of a watermark and is created to mislead the attacker who would try to crack the watermark typically as a watermark is normally broken. This is done in the view that an attacker, not realizing that a different element had been introduced, would be foiled.
We embedded a 2D Dataglyph as a watermark signature into a watermark. In case of symmetric watermarks, the management of the keys is another challenge. This problem is rectified by ensuring that a watermarker implements the key and only they alone know the key. The 2D DataGlyph will have its own key, whist the image maker will have the original image. Each will be separate and not shared amongst the others. The chances of all three keys being found are then minimal. Within these parameters we used ICA to see if it is possible to break the watermark.
The ICA technique extracts linearly independent distinctive parts from a data-set. Unlike decorrelating data methods, ICA searches for directions in data-space which are independent across all statistical orders [197]. ICA is capable of finding the underlying factors or sources when classic methods fail completely [198]. The approach reported here has the advantage of including a learning probabilistic model (as opposed to projections in data-space), while remaining computationally efficient in high dimensions.
Independent Component Analysis is a computational technique for unraveling a mul- tivariate signal into additive subcomponents under the assumption of mutual statistical independence of non-Gaussian source signals. It is a special case of blind source separation [200], [201] and [202]. ICA is a method of utilizing the process of computer power and statis- tical techniques for illuminating secret influences which are embedded within sets of random variables, measurements, or signals [197], [203]. ICA has the ability to examine data from various kinds of application fields, as well as digital images, document databases, economic indicators and psychometric measurements [198]. In numerous applications, the dimensions are rendered as a set of parallel signals or time series; the expression blind source separation is employed to differentiate this problem. Common examples are mixtures of simultaneous speech signals that have been picked up by several microphones, brain waves recorded by multiple sensors, interfering radio signals arriving at a mobile phone, or parallel time series obtained from some industrial process [199]. Here, we used the ICA to see if it is be able to
separate a 2D Dataglyph from a Watermark using the Delome model [204].
The data variables are linear mixtures of some unfamiliar latent variables, and the mix- ing system is also unfamiliar [204]. The hidden variables are non-Gaussian and mutually independent. They are called the independent components of the observed data [204]. The sources are mixed and then separated into two sources. Delorme [204] uses the time series of two independent sources A and B. Both sources are linearly mixed, i.e. added (or sub- tracted) to produce a linear combination of the form aA + bB where a and b are assigned coefficients. The ICA algorithm recovers the original sources A and B. Of import is the fact that the algorithm cannot recover the exact amplitude of the sources. Note that ICA can only extract sources that are combined linearly [199].
The focus of ICA is that it bestows unsupported grouping of multimedia data that has shown to be well-aligned with manual grouping [145], [144], [143], [142], [141], [140], [139] and [138] in different media.