Slashdot Mirror
Facebook To Fight Revenge Porn by Letting Potential Victims Upload Nudes in Advance (bleepingcomputer.com)
Catalin Cimpanu, writing for BleepingComputer: Facebook is testing new technology that is designed to help victims of revenge porn acts. It works on a database of file hashes, a cryptographic signature computed for each file. Facebook says that once an abuser tries to upload an image marked as "revenge porn" in its database, its system will block the upload process. This will work for images shared on the main Facebook service, but also for images shared privately via Messenger, Facebook's IM app. The weird thing is that in order to build a database of "revenge porn" file hashes, Facebook will rely on potential victims uploading a copy of the nude photo in advance. This process involves the victim sending a copy of the nude photo to his own account, via Facebook Messenger. This implies uploading a copy of the nude photo on Facebook Messenger, the very same act the victim is trying to prevent. The victim can then report the photo to Facebook, which will create a hash of the image that the social network will use to block further uploads of the same photo.
The idiocy of this is that if the revenge poster slightly alters the image (resize, re-compress, slight quality change, etc) it changes the hashing.
"All your nub are belong to us"
Oh. Another idiot. What about the ones others took, without you being aware
If someone else took the picture and the victim is unaware, the target probably doesn't have a copy to upload preemptively.
Comment removed based on user account deletion
There are much better methods of hashing images than stupidly taking a file checksum, such as this one here:
https://pippy360.github.io/tra...
This algorithm here does not care about affine transformations applied to the image, so it can be scaled, rotated, skewed, and still be a match.
Because English speakers know that "male" pronouns convey no information about gender. English is gender neutral unless female pronouns are used. So, when you want to make neutral statements you don't use female pronouns.
There are image similarity algorithms out there that do not care about the absolute hash of the file, and can detect the same image cropped, scaled, or rotated just fine.
Here is one such algorithm:
https://pippy360.github.io/tra...
Image hashes typically work in a way such that you can find an image even if minor alterations have been made, such as if you re-compress them, change formats, alter a single pixel, etc. From what I understand, it often involves analysis of the color histogram used in initial searches, plus a tiny thumbnail for direct comparison, which would generally be too small to recognize a specific person. This lets you do "fuzzy" matching, unlike a hash like CRC32 or SHA1 which only can find exact matches.
I agree that this has all sorts of psychological barriers. "Hey, I'm worried about revenge porn, so I'm going to upload all my nude pics I shared with my ex-boyfriend to Facebook for analysis. You know, Facebook, the company that scans all my personal data for profit."
Irony: Agile development has too much intertia to be abandoned now.
Also, in UK, Facebook could be charged with possession of child pornography and the teen uploading the photo with distribution