News for hackers. Stuff that matters.
Re:Inside Pornhub by willcipriano on Wednesday February 23, 2022 @10:45AM (#30442279)
I've never used it, but apparently you can hash a image in such a way that small changes will result in the same hash. I think that is what Apple was doing with it's child abuse filter announced earlier this year.
Re:Inside Pornhub by shakna on Wednesday February 23, 2022 @11:00AM (#30442475)
You're probably thinking of a perceptual hash. [0] It's a relatively new field, academically speaking, so only a few players dominate, and there is a huge amount of research going into it.

There are other techniques for automatic flagging of similar files, like PhotoDNA's [1] enormous database - which I believe PornHub are already using.

[0] https://en.wikipedia.org/wiki/Perceptual_hashing

[1] https://www.microsoft.com/en-us/photodna

Re:Inside Pornhub by toomuchtodo on Wednesday February 23, 2022 @11:35AM (#30442906)
It seems like a natural evolution of PhotoDNA to have this content (non consensual adult content) flagged programmatically, with a flow for folks to attest to content to be fingerprinted and scanned for.
Re:Inside Pornhub by mannykannot on Wednesday February 23, 2022 @10:58AM (#30442448)
In a sense, image recognition is like that - multiple different images hash to "cat", for example.

For the purpose here, robust hashing seems to be the relevant term:

https://www.netclean.com/netclean-report-2019/robust-hashing...

Re:Inside Pornhub by acchow on Wednesday February 23, 2022 @05:53PM (#30447740)
> I've never used it, but apparently you can hash a image in such a way that small changes will result in the same hash

This is equivalent to image recognition. It works ok, but has lots of false positives and false negatives.

Re:Inside Pornhub by tomatotomato37 on Wednesday February 23, 2022 @12:52PM (#30444010)
I don't know if perpetual hashes will work with full videos though, and linking the hash to a singular frame seems very unreliable, especially considering how randomly lossy video compression is
Re:Inside Pornhub by shakna on Thursday February 24, 2022 @01:40AM (#30451361)
Most video-based perceptual hashes work by taking a series of thumbnails at set intervals, say every 5% of the video, assembling a collage, and then making the perceptual hash against the collage.

As you note, it isn't particularly reliable, but it does actually tend to work fairly well against the loss of video compression. However, it doesn't tend to work well when the video is recut, with time added or removed. There is some research trying to combine approaches for finding important keyframes (ML-powered thumbnail searching) with perceptual hashing [0], which is looking promising, but it's still fairly unreliable.

[0] https://doi.org/10.1007/s10207-013-0211-z