Several people are attempting to create an algorithm to mask the presence of a steganized encrypted message in the least significant bits of an image. Don't forget that no matter how fancy your algorithm or how closely you mask your steganography with a model of what the statistics of an ordinary image look like, you have to assume that your opponent also knows your steganization algorithm, including your masking technique. (Otherwise you are just relying on security through obscurity.) This leaves you with three problems: (1) your opponent may have a much better model of an ordinary image than you do, and still be able to discern the existence of masked steganography, (2) since your opponent knows your steganization algorithm, he/she can look for any "signature" that your steganography masking model leaves, and (3) your opponent can "desteganize" all your images and check their statistics for deviations from the statistics for "desteganized" ordinary images. Resolving problems (1) and (2) requires a lot of work constructing good models. Resolving problem (3) requires, I think, a modeling function for steganography that is invertible only with a secret key. (Otherwise, your opponent could desteganize your image and find a uniform random distribution, which indicates an encrypted message.) Since this type of function is, to my knowledge, not well-developed, don't expect it to be secure. Thus, if breaking it could compromise your secret key for desteganization, then don't use the same public/private key pair for both encryption and steganography. Kevin Q. Brown INTERNET kqb@whscad1.att.com or kevin_q_brown@att.com