Slashdot Mirror
MySpace Begins Rollout of Video Monitoring Tech
C|Net is carrying an article looking into new technology MySpace is rolling out to combat user violation of copyright laws on their pages. Called 'Take Down, Stay Down', the service will attempt to ensure that once content is removed because of a complaint it can never be uploaded again. "Copyright owners have access to Take Down Stay Down free of charge, according to a release from MySpace. If the social-networking service receives a takedown notice regarding a copyrighted clip hosted through its MySpace Videos hosting service, MySpace's new feature will take a 'digital fingerprint' of the video and add it to a copyright filter that blocks the content from being uploaded again. '(It's) the ability to have a piece of content imprinted and put in a database so we can identify it,' said Vance Ikezoye, CEO of Audible Magic." The article goes on to discuss the problems YouTube is facing with the same issues, as well as recent investigations of this issue in the political arena.
Well, here's my idea. It's right off the top of my head, so it's probably patented:
To create the fingerprint:
1. Transform the frames to contrast edges. I forget what the filter is called but it'll basicly give you an outline of all the edges, with very little info
2. Split the frame into squares, calculate a brightness value from the edge frame
3. Throw all of this into a database
To check the fingerprint:
1. Decode a few keyframes
2. Calculate same info for those frames
3. Use a quick sort on brightness to find candidate frames. Don't require 100% match, make it so that a frame with a few watermarked squares will still score
4. Sanity check for file/location in other clip
5. Second pass either taking more frames, matching constrast lines or something, plenty opportunities here
Example, if they've blocked "mymovie.avi" with an NBC logo, and you're trying to match it against "othermovie.mpg" with a Fox logo:
1. Pick frames 154, 437, 1023, 2022 from Fox clip
2. Do the math
3. Find candidate matches
4. If they so *roughly* match frames 184, 467, 1053, 2052 in mymovie.avi you have a probable match with a 30 frame offset. In practise, just require an approximate match because of different commercial cut-offs
5. Verify match more closely
Sounds to me like a not too heavy job. It's very hard to fuck with because contrast lines are basicly the content of the picture. The same clip of a person walking will be approximately the same no matter how badly you fuck with it. Maybe you need to normalize it a bit if people try to upload high/low contrast vid to get around it, but that's going to look like shit anyway.
Live today, because you never know what tomorrow brings
I don't know if Audible Magic's come up with anything special for video, but for audio, the underlying technology is at least partly based on MFCC. You can find some more details here and here .
Also, they have to apply this filter only when content is added to a page, not every time it is played, so its a little less computationally intensive than some people have suggested.
It may be more expensive to search all existing fingerprints every time a new take-down request comes in.
fwiw - no, it catches that.
Most video fingerprinting technology can deal with mirroring, rotating, shearing, compression, time stretching, channel swapping (RGB and YUV), and practically any other method you can think of. One thing you -can- is chop it up into fragments that are smaller than the watermarking window, and distribute those fragments across the canvas randomly. The problem with that (and, actually, almost all of the aforementioned methods) is that the video is unwatchable and compresses horribly. The latter you're stuck with, the former you could code a special plugin for that unscrambles the video. But at that point, it's no longer video for the masses.
I'm sure other posters have already pointed out that people will just upload/download on a different service, etc. so I won't go into that here.. nor the possible (unlikely, but possible) 'fair use' issues.
Audible Magic software is one of the tools used by the RIAA "expert" in that recent RIAA suit where the RIAA expert was knocked down:0 2206:/
_ lindor_070223JacobsonDepositionTranscript:/
http://yro.slashdot.org/article.pl?sid=07/04/28/2
http://www.ilrweb.com/viewILRPDF.asp?filename=umg
Nothing like relying on trade secrets and black box algorithms to make you sure that you're taking down the right files and leaving up the clean ones.