I think the need to be OCR'd first. And that's a bit challenging given all the stamps and scribbles in there. Someone needs to train an AI to OCR only the underlying doc.