Steganalysis of LSB replacement for multivariate Gaussian covers

Recently, some statistically optimal steganalyzers are proposed based on hypothesis testing theory, in which the cover pixels are supposed to be independent. However, the independent assumption is of limited interest since redundancy exists in natural images. In this paper, using a more appropriate image model considering pixel correlation, a new steganaly-sis method for the least significant bit replacement (LSB-R) steganography is proposed. First, the cover image is divided into non-overlapping equal-sized blocks and each block is modeled by a multivariate Gaussian distribution. Then, based on likelihood ratio test and formula derivation, a new detector of LSB-R is derived. The proposed detector is an extension of some previous works with improved detection performance.