The basic idea
A camera or scanner captures an image of a page. To a computer, that image is just a grid of colored pixels, with no concept of "letters" or "words" built in. OCR software is the layer that looks at those pixels, figures out which groups of them form characters, and converts that visual information into actual text characters a computer can store, edit, and search.
The output is usually plain text, or text embedded invisibly behind an image in a searchable PDF, so the original page still looks the same but a computer can now find and copy the words on it.
Why it matters
Without OCR, a scanned document or a photo of a page is just a picture. You cannot search it, copy a sentence out of it, or have a computer read it aloud. OCR is the step that turns static images of text into usable digital content, which is why it sits underneath so many everyday tools: document scanning apps, digital archives, accessibility software, and pen scanners that read a line of text and speak or display it.
It is also the reason a scanned library book can show up in a search engine, or a photographed receipt can be turned into an expense line item automatically.
Where OCR shows up
OCR is embedded in many products people use without thinking of it as "OCR" specifically: phone camera apps that scan documents, PDF software that adds a searchable text layer, translation apps that read foreign-language signs, license-plate readers, and handheld pen scanners that glide over a line of text and convert it to speech or a digital copy in real time.
Each of these applications uses the same core recognition process, but tunes it for a different context, such as speed for a handheld device or maximum accuracy for archival scanning.
OCR versus just taking a picture
A photo of a page and an OCR-processed version of that page look identical to a human eye, but they behave very differently to software. A photo is an image file; a computer has no idea what words are on it. Run that same photo through OCR and you get actual text data, meaning you can select a sentence, search for a word, resize the font, or feed it to a text-to-speech engine.
A short history in brief
Early OCR systems in the mid-20th century relied on rigid template matching, comparing a character shape to a fixed library of known letterforms, which meant they only worked reliably on specific fonts. Over time, systems moved to feature-based matching that looked at strokes, curves, and proportions rather than exact shapes, making them more flexible. Modern OCR increasingly relies on neural networks trained on huge datasets of text images, which handle font variety and imperfect scans far better than older approaches.