[MPlayer-users] Subliminal image detection

Oliver Seitz info at vtnd.de
Thu Dec 9 22:57:59 CET 2010


>> Finding such pictures would be quite easy: [...] you can find
>> those pictures by just looking for two I-frames in direct
>> succession.
>
> A good encoder, at least for H.264, should not need more than
> one I-frame (note: not IDR, if it used that it would have to use
> two indeed).

Thanks for that information, I'm no video codec expert. I'm just a 
showoff. ;-)

> However your analysis does not take into account variable-frame-rate formats,
> which might change things a bit,

Of course regarding the time the frames are displayed, yes. And, if 
those images were used (which i doubt) I don't think they are already 
further obfuscated by encoding them by several very fast frames just not 
to exhibit two "scene changes" in three frames. And even if, then one 
could detect two scene changes within 0.1 seconds.

But still, the first publication on that subject claimed to have 
projected the subliminal pictures for 5 milliseconds. If such a frame 
time is possible with variable frame rate, to display it for 5 ms, you 
would have to have a 200Hz monitor refresh rate which is more than 
unusual. And, while a CRT could be able to display it, a TFT can not.

> and of course if you don't care about
> coding efficiency those frames can be P-codec as well.
> A scene change detection algorithm would be more precise, though much
> more effort etc.

Yes, while reading the manpage I briefly thought about suggesting to 
(ab)use the decimate filter, but that would require a lot more work for 
both the user and the CPU. Scanning for two consecutive I-frames works 
at very little CPU overhead while watching any video, and one could scan 
a lot of videos at about no effort at all.

Greets,
Kiste


More information about the MPlayer-users mailing list