How to OCR a scanned PDF on Windows

Does the PDF need OCR?

Try selecting a word or searching for a phrase you can see. If nothing is found and the whole page selects like one photograph, the PDF is image-only and needs OCR. A born-digital PDF can also contain damaged or incomplete character mappings, but blindly OCRing good text can introduce new errors. Use native text when it is reliable.

Common OCR candidates include:

• Documents scanned from a flatbed or office copier
• Phone photographs saved as PDF
• Fax archives
• Historical material and printed forms
• Mixed documents with image-only appendices

Choose the right OCR output

Extract text from an image without uploading it

If the source is a JPG or PNG rather than a PDF, first create a one-page PDF with Images to PDF. Choose Use original image size when preserving the photograph's proportions matters; choose A4 only when the output is meant to become a standard printed page. Open that PDF and use Recognize this page for a quick result, or To text when you need a TXT file. PDFCore sends the page to the OCR components already installed in Windows; it does not send the image to an online conversion service.

1. Prepare one representative image. Rotate it upright and crop empty borders, but keep the untouched original.
2. Create and open the PDF. For several photographs, put them in a deliberate order before conversion.
3. Select the document language. The chooser lists OCR languages actually available to Windows. A language that is not installed cannot be selected.
4. Recognize one difficult page first. Use a page containing small text, numbers, punctuation or multiple columns rather than an easy cover page.
5. Choose the deliverable. Copy the quick result, export TXT for plain content, create a searchable PDF to preserve the image, or create Word when editing matters.

A five-minute reproducible OCR check

Create a small test image containing a heading, one paragraph, a date such as 2026-08-21, a negative amount such as -1,234.56, and a two-column list. Run it once with the correct installed language. Then compare the output character by character and record errors in four groups: letters, numbers, punctuation and reading order. Repeat with the same image after rotating it slightly or reducing its resolution. This shows whether an apparent improvement came from the OCR setting or simply from a better source image.

Do not report a single “accuracy percentage” unless the counting rule and source are fixed. For practical checking, count the total characters in the reference transcription, substitutions, insertions and deletions, and keep the test image with the result. Names, account numbers, dates, decimal separators and minus signs deserve manual review even when the surrounding paragraph looks correct.

Create a searchable PDF

1. Open the source PDF. Preserve the original as an image record.
2. Choose OCR. Select “Make PDF searchable (recommended).”
3. Select the recognition language. The list is based on OCR languages installed in Windows. Choose the language used by most of the page text.
4. Save to a new filename. A suffix such as _searchable makes the result easy to distinguish from the untouched scan.
5. Let OCR finish. PDFCore recognizes image-only pages and adds the text layer without sending the document to a web service.
6. Search and select text. Test names, numbers, headings and a phrase from the middle of the document.
7. Compare appearance and page count. The searchable PDF should still look like the original scan because the image remains the visible page.

Improve OCR accuracy before recognition

Resolution

A sharp scan around 300 DPI is a strong general starting point for ordinary printed text. Very low resolution loses character detail; extremely high resolution increases processing and storage without guaranteeing better recognition.

Orientation and skew

Rotate pages upright. Even a few degrees of skew can make line and table detection harder. If a phone photo uses perspective, correct the trapezoid before OCR.

Contrast and background

Text should be clearly separated from the page. Remove dark scanner borders, shadows, show-through and uneven lighting where possible. Avoid aggressive thresholding that closes small counters inside letters.

Language

Choose the correct Windows OCR language. A mismatched language model can confuse accented letters, punctuation and word boundaries. Mixed-language documents may need separate passes or extra review.

Source typography

OCR is less reliable on handwriting, dot-matrix output, decorative fonts, tiny footnotes, mathematical formulas, vertical writing and low-quality photocopies. These need focused human verification.

Windows OCR language availability

PDFCore uses languages exposed by the local Windows OCR platform. If a needed language does not appear, install the corresponding Windows language features/OCR capability, restart the app, and check again. Language availability depends on the Windows installation and edition; PDFCore does not download a private cloud OCR model.

Convert a scan to editable Word

Choose “Convert scan to editable Word” or use PDF to Word and set the OCR language. Then select among three output goals:

• Editable layout: reconstructs paragraphs and tables while preserving graphics.
• Precise layout: keeps text placement closer to complex forms and photographed pages.
• Flowing text: simplifies layout for easier rewriting and reading order.

OCR does not know the original font, style names or table model. The Word output is an inference from pixels and coordinates. See the detailed format-preservation guide before processing a complex form or report.

Verify OCR where errors matter

Do not review OCR by merely glancing at the page image; a searchable PDF can look perfect while its invisible text layer contains mistakes. Test the recognized text itself:

• Search for a heading, a person’s name and a phrase containing punctuation.
• Copy a paragraph into a plain-text editor and compare it character by character.
• Check 0/O, 1/l/I, 5/S, decimal points, commas and minus signs.
• Verify dates, amounts, account numbers, legal citations and medication or part codes.
• Check column order by selecting text across the page.
• Inspect tables row by row rather than trusting visual alignment.

Troubleshooting

The page is searchable, but selection highlights the wrong area

The OCR coordinates may not align because of rotation, crop or unusual page scaling. Correct the page geometry before OCR and try the intended language again.

Words are recognized in the wrong order

Multi-column layouts, sidebars and tables can confuse reading order. For Word output, compare Editable and Precise layout. For plain-text extraction, plan to repair column boundaries.

Only some pages become searchable

Some pages may already contain native text while others are images. Test an image-only page directly. Check for very large, corrupt or protected pages and record the first page that fails.

The desired language is missing

Install the corresponding Windows language/OCR features, then reopen PDFCore. Availability is controlled by Windows.

Scan directly to a searchable PDF

With a compatible Windows Image Acquisition (WIA) scanner, PDFCore can choose 150, 200, 300 or 600 DPI, color/grayscale/black-and-white, flatbed or automatic document feeder, and duplex where the driver exposes it. Choose “Searchable PDF (OCR)” to create the scan and add text locally. Hardware and driver support determine whether feeder and duplex controls work.

Questions about OCR