If you're a regex wizard, that may be doable. Perhaps consider using Python instead? It'd probably involve less hair-pulling and be more maintainable by others.

In either case, please try it. Scraping the popular image hosts like Imgoat gallery individually would be one approach, or you can try to scrape Voat directly and back up every image that has at least one upgoat.

IPFS is great! I never heard about it until just now. One person can run a regular scraper script from imgoat (for example), and the rest of the peers will pull and replicate the images over time. Then a generic browser extension can be created that will attempt to pull dead links from IPFS. I bet their dev team is already working on something like that.