Steganographic applications of the nearest-neighbor approach to Kullback-Leibler divergence estimation

We propose to use a method for divergence estimation between multi-dimensional distributions based on nearest neighbor distance (NND) for optimization of stegosystems (SG) and steganalysis. This approach has previously been effectively applied for the purposes of estimation and classification (particularly in the field of genetics). However, since divergence (precisely speaking, Kullback-Leibler divergence) is very popular in steganography, the NND approach can be used in order to estimate the security (undetectability) of stegosystems, given the known cover object corresponding to the tested SG. We will show how affects on the estimated divergence methods of image embedding and their parameters. This allows optimization of SG in relation to it´s security for the given cover images. Stegosystem detection based on the NND approach is also considered.