What is OCR (optical character recognition)?

How OCR works

An OCR engine first cleans up the image: it may convert it to grayscale, raise the contrast and straighten it. It then finds regions that look like text, splits them into lines and words, and recognizes the characters. Modern engines use neural networks that read whole lines at once rather than guessing one letter at a time, which helps with unusual fonts.

The output is plain text, often with the position of each word on the image. For video, OCR runs on sampled frames, so the result also carries a timestamp for when the text was on screen.

An example

You recorded a two-hour conference and remember a slide titled “Q3 retention by cohort,” but nobody read the title aloud. The transcript can't find it. OCR on the video frames can, because the words were on screen. The same goes for a screenshot of an error message, a whiteboard photo or a street sign in a travel photo.

Screen recordings are a particularly good fit. Tutorials, product demos and recorded calls with shared screens are full of text that nobody says aloud: file names, menu items, code, error messages and spreadsheet figures. OCR turns all of that into something you can search, which is often the only way to find the moment a specific window or document was on screen.

Where OCR struggles

• Low resolution or motion blur. Small text in a compressed video frame may simply not have enough pixels.
• Handwriting. Printed text is much easier than handwriting, which many engines read poorly.
• Stylized text. Logos, heavy effects, curved or vertical text trip up most engines.
• Busy backgrounds. Text over a photo or pattern is harder than text on a plain background.
• Layout. Multi-column slides can come out in the wrong reading order.

OCR for video vs documents

Document OCR usually runs once per page and cares about layout, because you want a faithful copy. Video OCR cares about finding things: it samples frames, skips duplicates of the same slide, and ties the text to a time. Perfect spelling matters less there, since approximate matching can still find “retenton” when you search “retention.”

In MediaFind

MediaFind reads on-screen text in video frames and photos on your computer and makes it searchable next to transcripts and visuals. Search for a word from a slide, a caption burned into a video or a screenshot, and the result links to the exact frame. It is available on the free plan and runs offline. For a walkthrough, see search photos by the text in them or search screen recordings.

Frequently asked questions

Can OCR read text in a video?

Yes. The video is sampled into frames, OCR runs on each one, and the text is stored with the time it appeared, so a search can jump to that moment.

Does OCR work on handwriting?

Some engines handle neat handwriting, but accuracy is much lower than for printed text. Expect misses on quick notes and whiteboards.

Is OCR the same as image recognition?

No. OCR reads characters and words. Image recognition identifies objects, scenes or faces. Many search tools use both on the same picture.