What Does OCR Stand For? Optical Character Recognition

November 11, 2026 · 10 min read

What Does OCR Stand For? Optical Character Recognition Explained

If you're wondering what does OCR stand for, the answer is optical character recognition. It's software that finds letters, digits and symbols in an image and turns them into text you can search, copy and edit. Below you'll see how an OCR engine reads a page, what its mistakes look like, and how it differs from ICR, OMR and MICR.

What Does OCR Stand For, Word by Word?

Each word in the name does a job. Optical means the input is light: a scan, a phone photo, a screenshot or an image-only PDF. Character means the unit being read is a single letter, digit or symbol. Recognition is the matching step, where the software decides which encoded character each shape on the page most likely is.

The result is machine-readable text. Before OCR, a scanned contract is just pixels, so you can't search it, select a sentence or paste a figure into a spreadsheet. After OCR, the words exist as text, so all of that works.

The acronym has other meanings too. In UK schools, OCR is an exam board, originally Oxford Cambridge and RSA Examinations and now branded Cambridge OCR. In sport, OCR means obstacle course racing. This post is about the technology.

How an OCR Engine Reads a Page: Preprocessing, Segmentation, Recognition

Most engines work through the same five stages. We'll use Tesseract as the reference, because it's open source and its documentation explains each step plainly instead of selling a product.

  1. Capture. A scanner or camera produces a grid of pixels. Tesseract's docs say it works best on images of at least 300 dpi.
  2. Preprocessing. The engine binarises the image (converts it to pure black and white), removes specks of noise, and corrects skew. Tesseract binarises internally with the Otsu method, and its docs warn that a skewed page badly hurts line detection.
  3. Layout analysis and segmentation. The engine splits the page into blocks, columns, lines and words. Tesseract assumes a full page of text by default and offers other page segmentation modes for a single line, word or character.
  4. Recognition. Each word or line image is matched to the most likely characters. Older engines classify one character at a time, while Tesseract 4 and later read whole lines with a neural network.
  5. Post-processing and output. Dictionaries and language data nudge doubtful words toward real ones. Then the engine writes the result out. The Tesseract README lists plain text, hOCR (HTML), PDF, invisible-text-only PDF, TSV, ALTO and PAGE.

Formats like hOCR and ALTO keep the position of every word on the page, not just the words. That's how a viewer can highlight a search hit in exactly the right spot on a scan.

Is OCR an AI Tool? Pattern Matching vs Neural Networks

It depends on the engine, and on how loosely you use the word AI. OCR is far older than today's chatbots. Tesseract itself was developed at Hewlett-Packard between 1985 and 1994, open-sourced by HP in 2005, and developed by Google from 2006 until August 2017.

Early systems used pattern matching. They compared each character image with stored glyph shapes, or measured features such as strokes, loops and line endings. That's template matching, not machine learning in the modern sense.

Modern engines learn from data. Tesseract 4 added a neural network (LSTM) engine that recognises whole lines at once, while still offering the older engine that, in the README's words, works by recognising character patterns. Some newer document AI services go a step further, using large vision models that also interpret layout and meaning.

So the honest answer is a spectrum. Classic OCR is pattern recognition, current open-source OCR is machine learning, and a chatbot that reads your photo is a large AI model doing OCR as one of many tasks.

What Is OCR Scanning, and What Does an OCR Scanner Do?

A scanner is hardware, and it only captures pixels. The OCR part is software, running in the scanner's app, a PDF program or your phone's camera. So what is OCR scanning? It's a scan followed by a recognition pass that adds text to the file.

When you tick an OCR or searchable PDF option, the software usually keeps the page image as scanned. It then places an invisible text layer behind it, aligned word by word. You see the original page, but you can search it and select text. Tesseract's own invisible-text-only PDF output is that hidden layer on its own.

To do OCR scanning well, follow these steps:

  1. Scan at 300 dpi or more, with the page flat and straight.
  2. Use even light and avoid shadows from your hand or phone.
  3. Pick the document's language in the OCR settings.
  4. Choose searchable PDF if you need the original look, or plain text if you only need the words.
  5. Proofread names, numbers and dates against the image.

Phones do this without a scanner. Apple's Live Text, for example, picks up text in photos and the camera view. If you already have a scanned PDF and want Markdown out of it, our guide on converting PDF to Markdown covers that workflow. For photos and screenshots, our image to text converter guide walks through the built-in tools on each phone and computer.

How Accurate Is OCR? What Recognition Errors Look Like

Clean, printed, high-resolution text in a common font is the easy case. Accuracy drops with low resolution, skewed pages, decorative fonts, light text on dark backgrounds, multi-column layouts and tables. Tesseract's docs say plainly that it has a known problem with tables unless you add your own layout analysis.

Here's a short page from a delivery notice, laid out in two columns:

Order No. 1047                 Office hours:
The modern warehouse handles   Mon to Fri,
overnight orders for the       8am to 6pm
Clearwater branch.

And here's the kind of raw result a struggling engine returns:

0rder No. l047 Office hours:
The modem warehouse handles Mon to Fri,
overnight orders for the 8am to 6pm
Clearwater branch.

Each mistake has a cause. The capital O became a zero and the digit 1 became a lowercase l, because those shapes are nearly identical in many fonts. The letters rn merged into m, turning modern into modem, which is still a real word. Worst of all, the engine read straight across both columns, so the office hours landed in the middle of a sentence.

That last error is a segmentation failure, not a recognition one. Even with every word recognised perfectly, the order would still be wrong. Spell-checkers won't catch it, and neither will a quick skim.

No engine fixes a bad scan. Rescanning at a higher resolution beats any amount of cleanup, and you should always check numbers and names by eye.

What Is OCR in a Bank? OCR vs ICR, OMR and MICR

Banks use a whole family of recognition technologies, and the acronyms get mixed up. Here's how they differ:

TechnologyFull formWhat it readsTypical use
OCROptical character recognitionPrinted or typed textStatements, invoices, ID documents
ICRIntelligent character recognitionHandwritingHandwritten forms and amounts
OMROptical mark recognitionFilled bubbles and ticked boxesExam answer sheets, surveys, ballots
MICRMagnetic ink character recognitionSpecial digits printed in magnetic inkThe code line on cheques

The MICR line sits along the bottom of a cheque. It typically holds the bank code, account number and cheque number, printed in magnetic ink that stays machine-readable even when someone stamps or writes over it. Two MICR fonts dominate: E-13B in the US, UK, Canada and Australia, and CMC-7 in parts of Europe, such as France and Italy, and much of Latin America.

So in a bank, MICR handles the cheque's code line, ICR handles handwriting, and OCR reads the printed text on everything else.

From Screenshot to Markdown with Tesseract.js in Your Browser

You don't need to install anything to watch an OCR engine work. Tesseract.js wraps a WebAssembly port of Tesseract, so the same engine runs in a browser or in Node.js.

Our image to Markdown tool is built on it. It reads one PNG, JPG, WEBP or GIF at a time in 11 languages, and the image isn't uploaded to our server. It doesn't read handwriting, and it uses no AI model.

It also shows why segmentation matters. Raw OCR gives you flat lines, so the tool adds a layout pass on top. Short or ALL-CAPS lines turn into headings, list markers into Markdown lists, and wrapped lines into paragraphs. A screenshot of a short checklist comes out like this:

# SPRINT REVIEW

The demo build is ready and the release date holds.

1. Fix the login timeout
2. Update the onboarding copy

- Owner: Priya
- Due: Friday

That renders as a top-level heading, one paragraph, a numbered list and a bulleted list. Paste it into the editor and the live preview sits next to the Markdown, so you can proofread against the original screenshot.

We prefer local OCR for screenshots of anything private, such as account pages or internal dashboards, since no outside OCR service ever sees them. To keep the original picture next to the text, link it by URL with the syntax in our Markdown image guide. And if the text is headed for a chatbot prompt, Markdown suits AI tools better than flat text.

So, what does OCR stand for? Optical character recognition, and behind the letters sits a pipeline that cleans the image, finds the lines and guesses each character. Its errors follow patterns you can learn to spot. Try it on a screenshot with our free image to Markdown tool, then check the result side by side in the editor before you use it anywhere.