If OCR is available for free, I would like to deal with scanned PDF documents, which are especially common in single-column mathematical papers.