Steganalysis of stegostorage system

Abstract

The aim of this contribution is to detect the presence of messages
in JPEG images that were stored in the files using sequential embedding
into the least significant bits of DCT coefficients. When embedding,
the StegoStorage system was used. For steganalysis, we used and implemented
a process of calibration JPEG images, which makes it possible to investigate
a histogram from steganographic images which are similar to the histogram
of the original images without secret information. Also, the Pearson's
chi-square test of goodness of fit was implemented, whose results
indicate detectability of the presence of hidden messages through
images in the two databases, the first database containing 1000
JPEG images and the second containing 450 JPEG images. An interesting
result is that after saving messages into the images and compressing
the JPEG once more, the detectability decreased in this case by almost
half.

Moreover, this contribution provides a framework for further testing by calculating
the minimum square error (the weighted stego analysis), which also
indicates detectability in both databases, and after embedding the
messages into images and re-compressing them with different quantization
matrix, the detectability is only slightly reduced. Therefore, we
consider this method an appropriate one for further testing and utilizing
it when embedding into the LSB randomly, without filling the full
capacity of the carrier file.