If they can get that to work, I've got to say that's brilliant! I guess it would not have to "videoprint" in real time to scale, it simply needs ti run the algo as the video is uploaded.

There's going to be a whole new level of cloaking if they get that to work! :)

Sure, but isn't this pure propaganda again? The quarterly report telling "doing what we can".

...

If the video has some hidden value direclty recorded from the broadcast - like captioning, or showview codes, you know the code that tells your VCR/DVD when to start recording... or whichever - they can look for that. But if you re-encode a file to be so plain it hurts, that "fingerprint" or whatever gets lost.

I read the other day that not long after we brought up the possibility here on WW, they've started trying to identify shows by their audio stream. ( No connection though. ) That's a better, and a more credible idea. They've probably found out most of the shows by a simple speech recognition software that's available for like what... $200?

Video recognition ( as in an AI telling what / who's on the video ) is sci-fi. It's in the works, but no-one is there yet. Not even still images can be disassembled with a good accuracy, even though shapes are recognized, the emphasis of shapes within shapes is a problem. ... not to mention that with some creative lighting ( not all shows are recorded in studios ) it's not that easy to recognize shapes either.

And as the very same thread, which replied to the very same quarterly propaganda stated... it's not Google who's going to invest in a multi-billion space-age security and intelligence project. Not Google, not for DowntheTube, not to please stockholders that they're doing what they can to not get sued again and again.

...

they can track images so I don't see how this is imposible.

Impossible IMO, as soon as they implement such a system someone will be posting how to get around it on a video website. There's too many variables with video, you can encode the same video at ten different bitrates and each one is going to be unique with it's own display characteristics. You can add noise, remove noise... there's too many things you can do to change it.

One thing I do know for sure, this will require a tremendous amount of computing power. Analyzing a single image and trying to find a match is one thing but analyzing a sequence of video running at 30fps then trying to see if there is match is on whole other level.

If the video has some hidden value direclty recorded from the broadcast - like captioning, or showview codes, you know the code that tells your VCR/DVD when to start recording... or whichever - they can look for that. But if you re-encode a file to be so plain it hurts, that "fingerprint" or whatever gets lost.

As far as analog captures/recordings go those reside in the overscan area which looks like a band of noise on the outside edges of a video when you view it on computer, it's not visible on TV. Most people crop or mask this area for aesthetic reasons to begin with.

There might be some easy mechanisms by which end-users could deceive automated filters. However, I do not think end-users will go through the painstaking trouble of doing so. It's more likely they'll take their pirated videos elsewhere, which is good for youtube.

The few that do can eventually be weeded out through manual intervention.

Altering a video is not a painstaking process, actually maintaining original quality is painstaking if you don't know what you're doing :) In lot of instances it would just be the matter of changing a value in a process you alreay doing anyway.

I am a bit skeptical of the idea. But that doesn't matter. Let the MPAA put their money where their snouts are: if they think it's so easy to devise a "video fingerprint", let them write the code, and turn it over to Google. At that point, it would no longer be stone-cold-braindead-insane to ask Google what it would take to run the allegedly functional fingerprinting code on newly uploaded files.

It would, however, be insane for Google to admit that such an algorithm can exist: I'm sure they have hundreds of PhDs who know any number of reasons why it can't. And if Google admitted that such a thing could exist, someone might pass or construe laws to make them create it -- which would be expensive at best and ultimately impossible. (Remember, this isn't a static thing: it is orders of magnitudes harder than, say, than weeding out plagiarized spammy websites, all the while spammers are busy tweaking their sites to avoid yesterday's spam-detectors.)

It's the MPAA goons -- and a stupider pack of Luddites hasn't walked the planet since the agricultural revolution -- that claim this is technically possible. Let them prove it's possible by doing it. Until they've done that, there's no point in anybody, least of all a judge, paying any attention at all to anything they say.

Impossible IMO, as soon as they implement such a system someone will be posting how to get around it on a video website. There's too many variables with video, you can encode the same video at ten different bitrates and each one is going to be unique with it's own display characteristics. You can add noise, remove noise... there's too many things you can do to change it.

Absolutely nothing is impossible. Forget the bitrate and the detail and take a step upwards to the "viewed video". It's a collection of frames - images if you like with a soundtrack. A human could spot a copied video based on what he remembers and what he sees, teach the machine to do that and the problem is solved.

The prize is I think in being able to recognise the "general picture" not the small detail and if you recognise enough of the clip then the red flag goes up. To be sure, serious computing power may be required but for google I don't think that's an issue. It's just going to take some clever code and algos and that is also something they are good at creating (or buying).

Also, if you add noise or radically change the clip to cirumvent the system then you're basically destroying the end product so what's the point.

Well perhaps impossible is too strong a word, as you said nothing is possible. It's possible for me to win a couple million in the lottery tommorrow but I'm not holding my breath.

The prize is I think in being able to recognise the "general picture" not the small detail and if you recognise enough of the clip then the red flag goes up.

I'd argue that's where the issue lies, there is no detail. It's all "general picture" which is going to produce a lot of false positives.

You and someone else mentioned audio, that on the other hand may be practical. Especially if they used it to initially flag videos then see if they have video match.

You and someone else mentioned audio, that on the other hand may be practical. Especially if they used it to initially flag videos then see if they have video match.

Very true. Would be best to sample with a fairly coarse filter before applying too much computing time to the task... A couple of million in the lottery would also be more than acceptable ;-)

is it really that hard? say you converted all videos to high contrast black and white. you'd have black/white shapes moving around.

if you could compare the speed and general 'shape' to other banned videos wouldn't that flag up as a possible dodgy video?

then throw in the audio check

Sure it's that hard, let's take your example for instance. By simply cropping even a small portion of the video you have removed parts of your shapes and the shapes that remain are much larger relative to the frame size.

Flipping a video as if it were in a mirror is normally 100% watchable, especially if audio is flipped to match.