A new detector of LSB matching steganography based on likelihood ratio test for multivariate Gaussian covers

Recently, steganalysis based on hypothesis test theory becomes a focus. However, the correlation between adjacent pixels is not exploited and this is obviously not reasonable. In this paper, a new detector for least significant bit matching (LSBM) steganography is proposed with the consideration of pixel correlation. The cover pixels are modeled by multivariate Gaussian distribution and a new detector for LSBM is derived based on likelihood ratio test. Moreover, we propose to calculate the detector by considering only the smooth image pixels and this is very helpful for improving the detection performance.