Statistical Correlations and Machine Learning for Steganalysis.

In this paper, we present a scheme for steganalysis based on statistical correlations and machine learning. In general, digital images are highly correlated in the spatial domain and the wavelet domain; hiding data in images will affect the correlations. Different correlation features are chosen based on ANOVA (analysis of variance) in different steganographic systems. Several machine learning methods are applied to classify the extracted feature vectors. Experimental results indicate that our scheme in detecting the presence of hidden messages in several steganographic systems is highly effective. [ABSTRACT FROM AUTHOR]