Complexity of deciphering problem with respect to cover-text and secrete message ratio

To analyze the complexity of the problem, we test the algorithm on the some stegoscripts with different size of covertext. Using the same article Moby Dick, we generate 6 scripts with different covertext to secret messages ratio, from 2 to 7. We expect that as the size of covertext becomes larger, the deciphering problem becomes harder.

The Figure above shows the results with Z-score threshold 3 on all 6 scripts. As expected, the true positive rate goes lower as the covertext becomes larger, while the false prediction rate goes higher at the same time. However, even as the covertext is as 7 times big as the original article, WordSpy still did a decent job. It accurately predicted 75% of the article and only had 53% of its predictions false positive, i.e., one will be correct for each two predictions.