I have a wee bit of Stego experience as I've written a couple of implementations. Generally for it to be "undetectable", you shouldn't go with more than 25% of an image file, assuming 24-bit color, being data, as it quickly becomes apparent that there is something fishy going on. Your best bet is to create a kind of "keyed stegonagraphy" where you generate a series of keyed nodes, creating a cycle (in the graph theoretic sense) of nodes, each node corresponding to a pixel, and the entire cycle determined entirely deterministically from the key.
This is akin to key schedulers used in various cryptography schemes, I suppose. The idea is that you REALLY don't want to just shove your data all at the beginning of the file in order, as it becomes really easy to tease out the data with some cursory frequency analysis/bruteforcing. "Oh the first 20 pixels encode the first X bytes of <insert well known file type here>, BALEETED!"
Then you simply have each user pick their own key, stored locally, and have the cycle generated on the fly when encoding and retrieving data.
This is akin to key schedulers used in various cryptography schemes, I suppose. The idea is that you REALLY don't want to just shove your data all at the beginning of the file in order, as it becomes really easy to tease out the data with some cursory frequency analysis/bruteforcing. "Oh the first 20 pixels encode the first X bytes of <insert well known file type here>, BALEETED!"
Then you simply have each user pick their own key, stored locally, and have the cycle generated on the fly when encoding and retrieving data.