Statistical Inferences on Embeddings in Steganography

Abstract

This paper is concerned with topical problems in steganography on detection of embeddings and statistical estimation of the models parameters. Binary stationary Markov chains with known and unknown parameters are used as mathematical models of cover-sequences. ML-estimators for models parameters are constructed. Statistical tests for detecting embeddings are constructed based on run statistic, short run statistic and likelihood ratio statistic. For a family of contiguous alternatives the asymptotic power of tests based on run statistics is found. A polynomial algorithm is developed for the statistical estimation of the positions of embeddings. Two approaches for evaluating security of steganographic schemes are proposed