LLRT BASED DETECTION OF LSB HIDING

 K. Sullivan, O. Dabeer, U. Madhow, B. S. Manjunath, and S. Chandrasekaran
Dept. of Electrical and Computer Engineering
University of California at Santa Barbara,
Santa Barbara, CA 93106
1

Abstract

In this paper we consider a hypothesis testing approach
for detection of hiding in the least significant bit (LSB).
This steganalysis problem is a composite hypothesis test-ing
problem. We state a regularity condition on the image
histogram, which reduces this problem to a simple hypoth-esis
testing problem. We then develop a number of simple
practical tests based on the estimation of the optimal log
likelihood ratio statistic. We show that our tests signif-icantly
outperform Stegdetect, a popular hypothesis test
available in the literature. Our approach also leads to good
estimates of the hiding rate.