Hector Martin

unfortunately I do not generally have episode data, a large portion of images on powerlinesinanime are 3rd party submissions

however... I had an idea one day to make a crawler downloading every anime that exists and then indexing every keyframe

the main obstacle to making this work is tying videos (which I can trivially obtain in totality) to some sort of database

Does http://mrcn.st/t/hash.py  help? Gives you a 64bit perceptual hash per frame, hamming distance=similarity.

it's useful (thanks!) but computing similarity is the easy part; the hard part is tying random .mkv's to anidb ids

Sounds easy to me. Hash the file (ed2khash is the primary key) and look it up in the database. Most well known stuff is in there.