Your mistake is in thinking that computers "see the image", second, you somehow think the output of OCR is different from a PDF engine that renders it into structured data/text.