Implementation and comparison of data embedding algorithms

Abstract

Steganography is a field which focuses on embedding some secret message in some media. The two goals this Final Year Projects sets out to achieve are the creation of a taxonomy which classifies these steganographic algorithms by the specific methods in which the secret message is modulated with the chosen digital media. In the FYP, it was decided to focus on digital images as the cover media, and the secret message to be modulated consists of textual messages of varying length. The second goal of this project is to compare four of these embedding algorithms from the different categories in the taxonomy. The four algorithms compared are Least Significant Bit Replacement(LSBR), Least Significant Bit Matching(LSBM), a GraphTheoretic Approach to Steganography as proposed by S. Hetzl and P. Mutzel in [HM] using a technique known as a static minimum degree construction heuristic(SMD), and HUGO. The comparisons made between these algorithms will be between their encoding and decoding times and the distortion generated when creating the stego-image. The final comparison made is a detectability measure between these algorithms. An SVM classifier is trained and tested using the BOSSBase image dataset of cover images, and on the stegoimages generated by each algorithm type. From the experimental results, it can be seen that the HUGO algorithm is the superior algorithm out of the four compared.