... The graphic file is going to have a highly correlated structure, long runs of white space etc.
This is not the case for digital imagery. For any given band, if you sample a significant portion of the image it would not be surprising to see a more or less random distribution over the range.
The statistics for such a file would be different than the random distribution you'd get from using the wrong key.
If you decrypt with the wrong key, do you get a random distribution? Would this be the case for all wrong keys? Would the statistics change in any discernible pattern as the keys got "closer"?
Even if the graphics format is compressed, leading to a more even distribution, ...
This may give a better result. Most compressions involve a shift to the frequency domain with a quantization of the samples. You might be able to write a detector based on this discontinuity. You could also try throwing an edge detector or some other morphological recognizer at the output. It would still be cheaper than having to a man in the loop, but you're going to spend a lot of cpu time. jps -- Jack P. Starrantino (215) 674-0200 (voice) SEMCOR, Inc. (215) 443-0474 (fax) 65 West Street Road jps@semcor.com Suite C-100 Warminster, PA 18974