ability to compare the hash of a picture with a list

[JokerGermany]

Is your feature request related to a problem? Please describe. At the moment in several german rooms there is a person (probably from japan) who uses several matrix homeservers for creating a account, disguised as another user, and post offensive pictures.

Describe the solution you'd like mjolnir should be able to look up the hash value of a picture and if the hash is on the list, he should be able do delete the picture (and ban the user)

Describe alternatives you've considered Writing a Bot for this...

[bkil]

This might fend off a few unsophisticated script kiddies, but it could be worked around easily in an automated fashion just by resaving the jpeg (at another quality level or after adding any amount of fractional blurring, sharpening, color enhancement, etc).

I think proper solutions are too expensive to implement within the scope of a project that struggles to finish even the basics:

• https://en.wikipedia.org/wiki/Perceptual_hash
• https://en.wikipedia.org/wiki/Digital_video_fingerprinting
• https://en.wikipedia.org/wiki/Reverse_image_search#Algorithms
• https://en.wikipedia.org/wiki/List_of_CBIR_engines#CBIR_research_projects/demos/open_source_projects
• https://en.wikipedia.org/wiki/Content-based_image_retrieval#Techniques
• https://en.wikipedia.org/wiki/Automatic_image_annotation

Related: #181 #148

[Yoric]

This might fend off a few unsophisticated script kiddies, but it could be worked around easily in an automated fashion just by resaving the jpeg (at another quality level or after adding any amount of fractional blurring, sharpening, color enhancement, etc).

Yeah, my personal experience is that such filters just don't work.

I think proper solutions are too expensive to implement within the scope of a project that struggles to finish even the basics:

Harsh but not false.

I believe that perceptual hashes are a good idea but I don't think we'll have time to work on them in the near future. If you know someone with the will and time to work on it, I'd be happy to mentor and/or review, though!

[JokerGermany]

There is someone who is writing a bot for this issue at the moment. He says it's "chaos code" at the moment^^ https://github.com/Unkn0wnCat/matrix-veles

[turt2live]

Hash matching of any variety is best done at the server level rather than room level. Mjolnir has NSFW protections these days which does something similar, but the idea of hash matching is best done external to the bot.