CSI: Compressing Similar Images

I've got a 20 gig database of business pictures in JPEG format - they're closeup pictures of boat decks from 700 different boats. However many boats are similar, and many images share identical pixels from slightly different views. So I'm wondering if I can compress them with a differ compression engine that better exploits the similarity and overlap between images. I'm also wondering if I could fine tune a neural network image compressor.

What tools would you use in production? Or is this still experimental territory? And if so, which projects should I invest in?