Feedbacks from the designers of BA
 
Here is an interview I have registered today from Major Deakins and Captain Hale, the designers of the BOWS-2 system. I asked them questions about the situation that happened last week ...
 
 
Patrick:
Dear Major Deakins, dear Captain Hale, it seems that the "Broken Arrow" algorithm has encountered some difficulties during the first week. You know that we put lot of money (1)  to get the BA algorithm, can you comment for all the BOWS-2 participant about what really happened?
 
Deakins:
"I do appreciate the money that you and your associates have invested in this operation, but it IS an operation. It's a military operation. And you don't know a thing about that."
 
Hale:
Calm down Deakins, please. [Silence] Well, we wanted to be ready for the big War, I mean Episode 3, and not only for the first shots. So, at the very last moment, we decided to put yet another heavy shield around the target. But as it appeared during the first week, the shield was more an arrow attractor than a real shield.
 
Patrick:
Hum, I'm not sure to understand what you're saying ... Would you mind being a bit more explicit?
 
Hale:
Ok, here is the thing. Everybody knows that the Achilles' heel of watermarking detection is the lack of diversity in the watermark signal. It is easy to estimate the secret key when observing a lot of watermarked images because there is no hidden message that could modulate the mark. In order to render things harder for episode 3, we had to find a way to prevent a too easy estimation of the secret key ... and we decided to extract a would-be robust perceptual hash from each image and to diversify the secret key according to this hash. This was supposed to be the ultimate counter-attack since, instead of one key, several hundreds keys would have been to be estimated ...
 
Deakins:
Come on man, I was pretty sure we'd miserably fail using this hash thing ... but you didn't believe me!
 
Hale:
That's true, the hash was my idea. But, you didn't spot its weakness during your benchmark. I found it quite strange, until I discovered a bug in your simulation code. You were bypassing the hash! How could I have trusted you!
 
Deakins:
Relax, Hale! [Silence] And it happened what happened during the first week: it was very easy to break the shield. Several participants were able to prevent a correct detection because the hash was not robust enough... specially for the "Souvenir" image. The attacked image was such that the detection key was just different than the embedding key.
 
Patrick:
Ok I see, Major Deakins, can you explain us what you did this week-end?
 
Deakins:
We have just decided to remove the shield of course. The hash is no longer used, the keys are the ones that have been used during the embedding. "And now Everything is going smoothly, I assure you". Episode 3 will no longer be mission impossible, but just quite hard to be completed. We are ready, our target is fully protected for episode 1!
Finally, I'd like to say that we were surprised by the success during the first week, "I say god-damn what a rush !"
But now, "if you succeed, you will get a ton of it".
 
Hale:
you're thinking about the prize, right?
 
Deakins:
no, my personal consideration.
 
Hale:
the consideration of a guy who wrote buggy codes! That's appealing, sure!
 
Deakins:
Shut up, Mister perceptual hash!
 
Patrick:
Thank you both for sharing these informations. We hope that you are right and that we will not see 40dB in the hall of fame tomorrow!
 
Deankins:
Come on participants, sharpen your arrows, we are waiting for you!
 
 
(1): This is of course ironical.
Feedbacks from the designers of BA
Tuesday, July 24, 2007