New multilevel DCT, feature vectors, and universal blind steganalysis

Publisher Website
Google Scholar
Abstract
Universal blind steganalysis can detect hidden messages without using prior information about the steganographic system. Recently, Farid developed a wavelet coefficient, higher-order statistics based, universal blind steganalysis method. This approach is a global method which demonstrated a high-quality in performance standards. Fridrich and Goljan also presented a DCT based local targeted steganalysis method to break the F5 algorithm. However, both Farid"s and Fridrich and Goljan"s methods have some limitations. This paper presents a local universal steganalysis technique combining the advantages of both methods. The basic components of the presented method are: novel DCT multilevel decomposition with wavelet structure; a new set of feature vectors; and a modified kernel function in the Kernel Fisher Discriminant. Experimental results show the presented method offers better performance than commonly used schemes. Inherently, the presented method has the ability to localize the hidden information, it can capture stego information in small blocks, and it is functional using only a small training set.

This publication has 0 references indexed in Scilit: