Is it better to just render PDFs as images and OCR them rather than read directly then?