OCR explained: how scans become searchable text

OCR — optical character recognition — is the process of finding text in an image and converting it into machine-readable characters. A scanner produces a picture of a page. OCR tries to recover the words.

Without OCR, a scanned PDF is an image. You can look at it, but you cannot reliably select, search or copy the text. With OCR, the PDF can include an invisible text layer behind the image, so it looks the same but becomes searchable.

The basic pipeline

A typical OCR system:

1. Preprocesses the image: deskew, denoise, adjust contrast and remove borders.
2. Detects text regions, columns and lines.
3. Segments lines into words and characters.
4. Recognizes characters using a model trained on text.
5. Uses language context to correct ambiguous shapes.
6. Reconstructs reading order and layout.
7. Outputs plain text, a searchable PDF or structured data.

Modern systems often use deep learning for detection and recognition. The details vary, but the goal is the same: turn pixels into characters with useful positions.

Why scans are hard

Human reading is forgiving. OCR is not always.

• Low resolution blurs letter shapes.
• Skew makes lines diagonal.
• Faded ink reduces contrast.
• Speckles and stains look like punctuation.
• Handwriting varies enormously between writers.
• Complex layouts — tables, columns, footnotes — confuse reading order.
• Unusual fonts and mathematical notation are harder than body text.
• Mixed languages can confuse language models.

A clean, 300 DPI scan of printed text is close to ideal. A phone photo of a crumpled receipt is not.

Accuracy is not one number

A vendor may claim 99% accuracy. Ask: on what text, in which language, at what resolution, and measured how?

Character accuracy and word accuracy are different. A single wrong character can change a date, an account number or a dosage. For critical fields, always verify against the image.

OCR also struggles with:

• similar characters such as 0 and O, 1 and l, 5 and S;
• punctuation in small print;
• stamps and signatures;
• tables where cells are not clearly separated;
• text wrapped around images.

Searchable PDF vs plain text

A searchable PDF keeps the original page image and adds an invisible text layer. It preserves the look of the scan and makes it searchable. The text layer may not align perfectly, especially if the OCR engine misread the layout.

Plain text is easier to edit and process, but loses layout and images.

Structured output such as JSON or XML can include positions, confidence scores and table structure. It is useful for automation but requires more work to use.

When to use OCR

OCR is useful for:

• digitizing paper archives;
• making scanned contracts searchable;
• extracting data from invoices and receipts;
• preparing a scanned PDF for PDF to Word;
• accessibility and text-to-speech workflows.

It is not a substitute for keeping the original. Always retain the source scan, especially for legal, medical or financial records.

Privacy matters

OCR can run locally in a browser or on a device, or it can run on a server. Uploading a document to a free online OCR service means trusting that service with the contents.

For sensitive documents, prefer:

• offline OCR software;
• a local browser tool;
• a provider with a clear retention and deletion policy;
• your organization's approved document workflow.

How to improve results

• Scan at 300 DPI or higher for normal text.
• Use grayscale or color as appropriate; avoid overly aggressive black-and-white thresholds.
• Keep the page flat and straight.
• Crop out dark borders.
• Choose the correct language before recognition.
• Split multi-column pages if reading order is wrong.
• Review names, numbers, dates and amounts manually.
• Run OCR again with a different engine if the first result is poor.

OCR and PDF to Word

Converting a scanned PDF to Word is a two-step process: OCR first, then export the recognized text. A plain “PDF to Word” converter that only extracts text will return nothing useful from a scan.

Our PDF to Word tool is for PDFs that already contain selectable text. For scans, run OCR first, then convert the OCR'd PDF. For compression, see how to compress a PDF.

The short version: OCR is powerful, but it is recognition, not understanding. Treat the output as a draft and verify anything that matters.