Fair warning: I work at Diffbot.

Essentially that's what Diffbot (https://www.diffbot.com/) does, except we don't the render pages as an image nor do OCR.

Diffbot renders the page in a headless browser, and uses computer vision to automatically identify the key page attributes and extract normalized data for specific page types (Articles, Products, Discussions, Profiles, Images, and Videos).

This approach enables us to work in any language and on sites that we've never come across before automatically with better than human level accuracy.