The current implementation by John Murdock is available here: https://hg.codeplex.com/forks/jrmurdoch/imageprocessing
This is the almost final design as it has been extensively discussed offline, and it's working fine right now. Definitely something we want in core. I might do a demo at the weekly meeting so we can discuss further what needs to be done next.
Sure thing is about listing all the filters we want to provide by default, and what APIs should be used for that, taking licence constraints into consideration.