Miscellaenous Steganography Techniques
There are numerous steganographic techniques and there it is not feasible to go through
all of the. Apart from the most popular ones covered in the main sections of this document
here are some of the
Spread-spectrum models
This is very popular for watermarking (see watermarking section). The steganography
applicaton is similar in that the cover image is the wideband signal and the
message is the narrowband signal and the message is spread over the perceptually significant
areas of the cover image (usually the top DCT coefficients)
Quantization and Dithering
In each quantization step a quantization error is introduced. For highly correlated signals
we can expect the difference signal to be close to zero, so an entropy coder can be used
effectively here. For steganography , the quantization error in a predictive coding scheme
can be utilized; that is we adjust the difference signal so that it transmits more information.
This scheme is morefully described in [4]
Information hiding in Binary Images
Binary images like fax data contain redundancies in the way black and white pixels are distributed.
Simple LSB schemes dont work too well for these images as they are highly susceptible to transmission
error and hence arent robust. Zhao and Koch suggest using the number of black pixels in a specific
region to encode secret information[13]
Statistical Steganography
This uses the existence of the "1-bit" steganographic scheme which embed one bit of information
into a digital carrier. This is done by modifying the cover in such a way that some statistical
characteristics of the change significantly if a "1" is transmitted; otherwise the cover is
unchanged. So the receiever must be able to distinguish between modified and unmodified covers
Distortion of digital images
Using a similar approach as substitution systems, the sender first chooses l(m) different
cover-pixels he wants to use. To encode 0 sender leaves the pixel unchanged, for 1 he adds
a random value to the pixel color. Though this resembles LSB there is no cover modification
if the data is 0 and due to addition of random number if 1 the statistical properties of
the image are better preserved