Video OCR
Video OCR (optical character recognition) is the process of detecting and reading text that appears on screen in a video, such as burned-in captions, slides, signs, scoreboards, name tags and product labels. The system samples frames, finds text regions and converts them to machine-readable text with timestamps. The result makes on-screen words searchable even when nobody says them aloud.
Why it matters
A lot of meaning in video is written, not spoken: names on lower thirds, slide text, prices, scores and signs. Without OCR, none of it is searchable. Reading on-screen text improves search, tagging and moderation, including catching scam links and phone numbers in overlays.
How it works
- Frames are sampled from the video, more often when on-screen text changes quickly.
- A text detection model finds the regions of each frame that contain words.
- A recognition model reads each region and converts it to machine-readable text.
- Repeated text across frames is merged, and each line is stored with its timestamps.
How do I extract text from a video?
Use a video OCR tool or API that samples frames, finds text and returns it with timestamps. For a quick one-off job, you can also export still frames and run them through an image OCR tool. Automated video OCR suits libraries better because it merges repeated text and records when each line appears.
What is the difference between video OCR and speech-to-text?
Video OCR reads words that appear on screen, such as slides, captions and signs. Speech-to-text converts words that are spoken aloud. They capture different information, so many indexing systems run both. For example, a lecture's slide titles come from OCR, while the explanation comes from speech-to-text.
Where it's used
News teams use video OCR to capture names and places from on-screen graphics. Online learning and SaaS training tools extract slide text so lectures are searchable, and creator platforms check text overlays for scams.
