How to convert a scanned PDF to Markdown with OCR

To convert a scanned PDF to Markdown, add it to MD Factory. The app finds the pages that are only pictures of text, recognizes the text on your PC with built-in OCR, and saves a Markdown file next to the PDF.

OCR handles clean printed pages well and makes mistakes on faded, crooked, or handwritten ones, so plan to check the result against the scan. The steps below cover what differs from an ordinary PDF: the OCR language, the Needs review status, tables, handwriting, and scanner settings.

Convert a scanned PDF

1. Open MD Factory.
2. Click Add files and choose the PDF, or drag it onto the window. To convert a whole folder of scans, including its subfolders, click Add folder.
3. Conversion starts right away. For the first PDF, the row reads Loading the PDF engine before the pages are processed.
4. Wait until the row shows the result: Completed, Needs review, or Failed.
5. Double-click the result path in the row to open the Markdown file.

The Markdown file gets the name of the PDF and is saved in the same folder. If the pages contain pictures, such as photos or diagrams, they go into a folder beside it. The PDF itself is not changed.

Scans/
  lease-agreement.pdf
  lease-agreement.md
  lease-agreement_images/

PDFs are converted one after another, while other document types in the queue run alongside them. The app reads files on local drives only. If your scanner saves to a network folder, copy the PDFs to your PC first.

How the app decides a page needs OCR

A page needs OCR when its text exists only as a picture. MD Factory checks every PDF for selectable text and chooses one of two ways:

• If no page has selectable text, the whole file is treated as a scan, and every page is recognized with OCR as a whole.
• If some pages have text, the app keeps that text as it is and runs OCR only where text appears as an image, such as a scanned page inside a typed report.

A mixed PDF needs no preparation. In our test, a three-page file with a typed first and last page and a scanned middle page came out as one Markdown file with the text of all three pages.

MD Factory takes scans as PDF files only. Image files such as JPG, PNG, or TIFF are listed as skipped, so set your scanner app to save PDF.

If your PDF already has selectable text throughout, OCR plays no part. The guide to converting PDF to Markdown covers those files.

Choose the OCR language

Every PDF starts with the OCR language set to Auto (script detection). In this mode, the app works out which alphabet each page uses, such as Latin, Cyrillic, Greek, or Arabic, and reads the page with a model for that alphabet. In our tests on clean test pages at 300 dpi with 12-point text, Auto read an English page without errors and a Russian page with fewer than one wrong character per hundred.

You choose a language when a PDF finishes as Needs review or Failed:

1. When the queue has finished, click the status in the file's row. A window opens with the list of problems found.
2. Open the OCR language list and pick the language of the document.
3. Click Retry.

There is no setting for choosing the language in advance. Each new PDF starts in Auto, and the language you picked last is preselected the next time you retry.

Two languages on one page work in Auto when they share an alphabet, such as English and German.

When the alphabets differ, Auto reads the whole page as one of them. For most alphabets, that model also covers English, so English terms on a Greek or Arabic page usually come through. Cyrillic is the exception: in our test, a page with both English and Russian text got Needs review. Retry with the page's main language, then check the words in the other alphabet against the scan.

What Needs review means for a scan

Needs review means the app saved the Markdown file but found a problem worth checking. For scans, the window that opens from the status usually shows one of these messages, with the page number in parentheses:

If no page yields any text, the file is marked Failed, and the window suggests retrying with an explicit OCR language.

Completed does not mean every character is right. The check looks at the average for a whole page, so a single misread character on an otherwise clean page does not trigger it. On the clean Russian test page mentioned above, the number 1248 came out as 124%.

Before you rely on the text, compare every figure you will use with the scan: names, dates, amounts, and account or ID numbers.

Crooked, noisy, and faded pages

A slight tilt and some specks are not a problem. In our tests, a test page tilted by about one degree, slightly blurred, and covered in fine specks converted as Completed.

A poor scan gets flagged. A blurred, low-contrast page scanned at low resolution and tilted by 3.5 degrees got Needs review.

Rotate pages that lie sideways or upside down in your scanner app or PDF editor before converting. When a page gets Needs review because of its quality rather than its language, scan it again with the settings below.

Tables in scanned pages

MD Factory looks for tables on scanned pages too and writes them as Markdown tables. Every cell is still OCR text, and a table hides mistakes well: a wrong digit in a column of figures looks as plausible as a correct one.

Compare the table with the scan row by row. Add up a column and check the result against the printed total. If values have moved to the wrong row or column, fix them in the Markdown. The PDF guide has an example of restoring a table that lost its rows.

Handwriting

MD Factory recognizes printed and typed text. It is not made for handwriting: handwritten notes, comments in the margins, and filled-in form fields can come out garbled or go missing.

Do not rely on the converted text for anything handwritten. Keep the scan as the reference and type those parts yourself.

Scan for better OCR

Recognition quality depends on the image. Tesseract's guide to improving OCR quality lists what makes an image hard to read, and for scanning it comes down to five things:

• Scan at 300 dpi or more. The guide says recognition works best at a resolution of at least 300 dpi.
• Lay the page straight on the glass. A tilted page makes it much harder to pick out the lines of text.
• Keep the lighting even, especially when you photograph pages with a phone. Shadows and an unevenly dark background make it harder to tell text from paper.
• Close the scanner lid or crop the edges. Dark borders around a scanned page can be read as extra characters.
• Use the cleanest copy you have. Specks and other noise lower accuracy.

Save each document as its own PDF. MD Factory creates one Markdown file per PDF, so a stack of unrelated papers scanned into one file becomes one long note.

IDs, contracts, and medical records

Scans are often personal: passports, contracts, bank statements, medical records. MD Factory recognizes them on your PC and does not upload the PDF or the recognized text.

The Markdown file is saved next to the scan. Text that was locked in a picture is now plain text that anyone with access to the folder can read and search. If that folder syncs to OneDrive or another cloud service, the Markdown file syncs too. Our guide to converting documents without uploading them shows how to check where a folder is stored.

For legal, medical, and financial documents, the scan remains the original. Treat the Markdown as a working copy, and check every number against the scan before you act on it. Delete the Markdown file and its images folder when you no longer need them.