What Is OCR? How to Extract Text from Scans and PDFs

OCR converts images of text — scans, photos, screenshots, image-only PDFs — into machine-readable characters. The distinction matters most with PDFs, which come in two kinds that look identical: digitally created PDFs (exported from Word, for instance) already contain real text you can select, while scanned PDFs contain only a photograph of each page. The quick test: try to select a word. If you can't, and searching finds nothing, you need OCR.

OCR output typically takes one of two forms: plain extracted text you can paste anywhere, or a 'searchable PDF' where an invisible text layer is placed behind the original scan — the page still looks like the scan, but search, select, and copy all work.

Classic OCR matched character shapes against templates; modern engines are neural networks. Tesseract — the open-source engine that powers a large share of OCR tools, originally developed at HP and later maintained by Google — switched to an LSTM neural network in version 4, reading whole lines in context rather than isolated characters. The pipeline looks roughly like this:

1. 1

   Preprocessing

   The image is converted toward clean black-on-white: deskewed (rotation corrected), denoised, and binarized. Most recognition failures are really preprocessing failures.

2. 2

   Layout analysis

   The engine segments the page into blocks — columns, paragraphs, tables, images — and determines reading order.

3. 3

   Recognition

   A neural network reads each text line and predicts character sequences, using language models to resolve ambiguity ('cl' vs 'd', 'rn' vs 'm').

4. 4

   Output assembly

   Recognized text is assembled with position data, producing plain text, structured formats, or a searchable text layer for a PDF.

On a clean, well-lit scan of printed text, modern OCR routinely exceeds 99% character accuracy. On a skewed phone photo of a crumpled receipt, it can fall apart. The factors that matter, in rough order:

• Resolution: ~300 DPI is the standard recommendation for scans. Below ~200 DPI, characters lose the detail engines need; far above 400 DPI adds little.
• Contrast and lighting: dark, even text on a light background. Shadows across a phone photo are a top cause of garbage output.
• Skew and perspective: straight-on scans beat angled photos. Many scanning apps auto-correct perspective — let them.
• Fonts and layout: standard printed fonts recognize nearly perfectly; dense tables, multi-column layouts, and stylized type are harder.
• Language settings: telling the engine the document's language lets its language model fix ambiguous characters. Good tools support dozens — FileMorf's OCR supports 14, including English, Spanish, French, German, Japanese, and Chinese.
• Handwriting: a different problem (often called ICR). Neat print handwriting partially works; cursive generally needs specialized services and still disappoints.

Before reaching for any tool, know what's built into your devices — for quick one-off grabs, these are excellent:

• iPhone/iPad and macOS — Live Text: select and copy text directly inside photos, in Safari images, and in Preview. Point the camera at a document and copy from the viewfinder.
• Windows — PowerToys Text Extractor: press a shortcut, drag a rectangle over anything on screen, and the text lands on your clipboard. The Snipping Tool in Windows 11 has a similar 'text actions' feature.
• Android — Google Lens: built into the camera and Google Photos; select, copy, or translate text in any image.
• Google Drive: upload an image or scanned PDF, open it with Google Docs, and Drive runs OCR automatically — a capable free option if uploading the document to Google is acceptable.

Their common limits: they're built for a screenful of text at a time, they don't batch-process multi-page PDFs well, most don't produce searchable PDFs, and layout (columns, tables) is frequently scrambled.

For a 30-page scanned PDF, you want a proper document OCR flow rather than screenshot-grabbing. FileMorf includes one built on the Tesseract engine:

1. 1

   Open the OCR tool and sign in

   OCR is a workspace feature: free accounts include 20 OCR pages per month. Unlike FileMorf's image and PDF tools — which run entirely in your browser — OCR jobs run in secure cloud processing, with outputs retained in your workspace and downloadable via rotating links.

2. 2

   Upload your scan, image, or PDF

   Pick the document language from the 14 supported — this meaningfully improves accuracy on non-English text.

3. 3

   Review and export the text

   Results open in your workspace, where you can review the extracted text, copy it out, or keep it alongside the source for later comparison.

• Making scanned archives searchable: a filing cabinet's worth of scanned PDFs becomes a searchable library.
• Receipts and invoices: extracting amounts and dates for expense reports and bookkeeping.
• Digitizing printed matter: pulling quotes from books, converting old typewritten documents, rescuing text from faxes.
• Data entry from forms: turning stacks of paper forms into rows in a spreadsheet (pair the output with a CSV conversion).
• Accessibility: screen readers can't read image-only PDFs; adding a text layer makes documents usable with assistive technology.
• Translation: text must be machine-readable before it can be machine-translated — OCR is step one for foreign-language documents.