Estimation of MB steganography based on least square method

In the available Jpeg steganography methods, model-based (MB) steganography technique is more secure than Jsteg, F5 and OutGuess. In this paper, we consider the problem of estimating the embedded length of secret messages of MB steganography. We provide an expected distribution of the non-zero AC coefficients´ high precision histogram to fit the stegotexts´, and adopt the cropped and recompressed calibration to obtain the carrier image approximation. Then, we present a new algorithm to estimate the embedding rates based on least square method. The attacks towards MB are successful with experimental evidence on 2700 carrier images and the stego images of different quality factor and different embedding rates.