I am talking about something-you-know identification factor. Examples of those formats are passwords, passphrases, pin-s, picture passwords ( http://blogs.msdn.com/b/b8/archive/2011/12/16/signing-in-with-a-picture-password.aspx ).
I am talking about the actual entropy that users need to remember, so "mothers maiden name" is weaker than random string of characters. About http://xkcd.com/936 : if there are 10k common words, then there are (10k pick 4) possibilities, so it's not significantly more entropy than an 8-character password over all 95 ASCII chars (95^8), once the method gets popular.
My question is based on the assumption that different formats of same information (sound, image, text, song, etc.) are not equally easy to remember.
So, what is the best format in easy-to-remember / entropy tradeoff? I'd be glad if you could link some relevant research too.
EDIT: I don't want to restrict the question to only text-based passwords, but rather include all human senses and computer input methods.