Steganalysis Based on Fast Sparse RepresentationClassification on Least Significant Bit Steganography

Detection of hidden message existence in low-rate steganography is still a challenging problem in manysteganalysis systems. The recently proposed steganalysis scheme based on sparse representationclassification has shown relatively remarkable detection capability in low embedding rate. In this paper,we propose a new steganalysis system using a fast sparse representation classifier. We compare ourproposed method with other steganalysis systems which uses different classifiers, including nearestneighbor, support vector machine, ensemble support vector machine, and conventional sparserepresentation based classifier on LSB steganography method. In all of our experiments, input features tothe classifier are kept intact to examine the ability of classifier to discriminate between innocent and stegoimages.

Keywords:

Steganography , Steganalysis , Dictionary Learning , Fast Sparse Representation