Very nicely done, and it looks like a totally manual process still? I was just working on processing some old beekeeping book illustrations the other day, using scans from the Internet Archive. Based on Mike Bostock's article /Why Use Make/ [1] where he explains the usefulness of capturing your process in a makefile for reproducibility, I made a makefile that downloads the source imagery, crops, adjusts levels, sharpens, and outputs PNGs of just the illustrations [2].

After doing all this manually, I found a writeup by Chris Adams where he talks about a process for using computer vision to automatically extract figures from pages of text [3]. So that's my current side-project.

Finally, after all of this, I was searching the Flickr Commons for imagery and noticed that the Internet Archive already has gobs of book illustrations extracted and posted to Flickr [4]! There's so many it must be an automated process, but I haven't found any details. They don't seem to be uploaded with the best quality possible, and the captions aren't included, so I think I'll continue on my quest (which is currently focused on generating high-quality public domain beekeeping-related imagery).

[1]: http://bost.ocks.org/mike/make/

[2]: https://github.com/beardicus/bk-fig-phillips

[3]: http://chris.improbable.org/2013/08/31/extracting-images-fro...

[4]: https://www.flickr.com/photos/internetarchivebookimages/

edit: formatting