The Result of Steganography Detection Test by Stego WatchTM and our Assessment of the detection program
(KIT STEGROUP)
The result of the detection experiment is summarized in the table bellow. The output of the detection program (decision result) is described by the symbols HP, P, and N.
| HP | High possibility to be a stego image |
| P | Some possibility to be a stego image |
| N | Not a stego image |
| Sample Image | Decision Result | Evaluation Score |
| sample-x1 | HP | -2 |
| sample-x2 | P | +1 |
| sample-x3 | N | +2 |
| sample-x4 | N | +2 |
| sample-y1 | HP | -2 |
| sample-y2 | HP | +2 |
| sample-y3 | N | -2 |
| sample-y4 | N | +2 |
| sample-z1 | N | -2 |
| sample-z2 | N | +2 |
| sample-z3 | HP | -2 |
| sample-z4 | N | +2 |
| sample-z5 | P | +1 |
| sample-z6 | P | -1 |
| sample-z7 | HP | -2 |
| sample-z8 | N | -2 |
| Total Evaluation Score | -1 | |
Only the file names in red are secret-data-embedded image (stego image)
Evaluation Scores in this table are determined by the following chart.
The maximum value of the "Total Evaluation Score" in this calculation amounts to +32 (the case all decisions are correct), and the minimum value -32 (the case all decisions are wrong). Actually, it was -1 in this experiment.
This tells us that this detection program did not provide any reliable information to us. In other words, we knew that this program couldn't cope with our Qtech Hide & View. We believe this is a case that the steganalyst didn't know the embedding method (Qtech Hide & View method).
We can also conclude that Stego WatchTM couldn't detect BPCS-Steganography, because the summation of the evaluation scores for sample-y1 - samply-y4, and sample-z1 - sample-z4 (which are either bmp, or png files) is just 0.
We hope someone working on steganalysis can analyze our sample images and report us the result whatever detection program he/she uses. Please go to the Formed Message sending page when you contact us.
We also have a set of specially embedded image files on the other page. If you have a steganalysis program, just try to analyze them to test your program. We are very much interested in hearing from you.
(Updated on April. 03, 2009 by Eiji Kawaguchi)