A Practical Method to Determine Achievable Rates for Secure Steganography

With a chosen steganographic method and a cover image, the steganographer always hesitates about how many bits should be embedded. Though there have been works on theoretical capacity analysis, it is still difficult to apply them in practice. In this paper, we propose a practical method to determine the appropriate hiding rate of a cover image with the purpose of evading possible statistical detections. The core of this method is a non-linear regression, which is used to learn the mapping between the detection rate and the estimated rate with respect to a specific steganographic method. In order to deal with images with different visual contents, multiple regression functions are trained based on image groups with different texture complexity levels. To demonstrate the effectiveness of the proposed method, estimators are constructed for selected steganographic algorithms for both spatial and JPEG transform domains.