OCR (text recognition)

Snipdeck reads the text inside your screenshots so you can search for it later and copy it as plain text. Recognition happens on your computer. Snipdeck downloads no model and does not go online to read text.

This page explains which engine each platform uses, how background reading works, how the recognized text feeds search, and how the OCR + Clipboard action copies a snip’s text.

What OCR does in Snipdeck

OCR (optical character recognition) turns a snip’s pixels into text. Snipdeck uses it for two things:

• Search: every snip you capture is read in the background and its text is stored. Later you find the snip by typing words that appear in it.
• Copying: right-click a snip to copy its text to the clipboard, then paste it into an editor, a chat or a terminal.

To read a snip’s text in another language instead of copying it, see OCR + Translate.

The engine

Snipdeck ships no OCR model of its own; it uses the engine on your system.

The snip’s right-click menu shows which engine runs, next to OCR + Clipboard.

Windows

On Windows, the languages OCR can read depend on the language packs installed in Windows. Snipdeck first asks for an engine for your profile languages, so an installed English or Turkish pack is picked automatically. If your profile has no OCR-capable language, en-US is used.

Linux

On Linux OCR is optional. Without Tesseract, Snipdeck still captures and stores snips; it just skips reading their text, and nothing fails. Install Tesseract and its language data with your distribution’s package manager. On Debian and Ubuntu, for example:

sudo apt install tesseract-ocr tesseract-ocr-tur

Snipdeck hands the snip to Tesseract through a pipe and writes no temporary image to disk, so the snip never sits in a file other users could read while recognition runs.

On both platforms, very small images (shorter side under 40 pixels) are scaled up before recognition; a single line of interface text reads better that way. Recognition runs on a separate thread, so the gallery does not freeze meanwhile.

Background reading

You never need to start OCR yourself. As soon as a snip is captured:

1. The snip is saved to the gallery and shows immediately; capture never waits for OCR.
2. A background thread reads the snip’s text.
3. The text is written to the snip’s record and added to the search index. The card gets the OCR badge.

Because this runs in the background, a new snip becomes searchable by its text a moment after it appears. Captures without readable text (a flat colour, a photo with no lettering) produce no OCR text. The snip is still saved and remains searchable by its other details, such as the window title.

Gallery search

The recognized text is kept with each snip’s other details in a local SQLite index. When you type in the gallery’s search box, your query is matched against all of these:

You can find a screenshot of an error dialog by a word in its message, or a snip of a web page by a phrase on the page. For matching rules and filters, see Gallery and search. The Has OCR and No OCR options of the Content filter separate snips whose text has been read from those whose text has not.

OCR + Clipboard

To get the text of one snip, right-click a floating snip or a gallery card and choose OCR + Clipboard.

• If the snip’s text has already been read, the stored text is copied to the clipboard at once.
• If not, Snipdeck reads the image on the spot, stores the result and copies it.

Either way the text is ready to paste, and the snip’s stored text is updated too. If no text is found in the image, nothing is copied.

It helps wherever you cannot select the text: a code snippet on a remote desktop, an address in a video frame, an error message in a PDF that arrived as an image. To get the text translated rather than as it is, use OCR + Translate; see OCR + Translate.

Privacy

Capture and OCR are local. Windows.Media.Ocr on Windows and Tesseract on Linux run on your computer; the contents of your snips are not sent anywhere while they are read, searched or copied. The only OCR-related feature that uses the network is OCR + Translate: it sends the recognized text (never the image) to Google Translate, only when you run it and only after you have agreed on first use. See Privacy and security for details.

See also

• OCR + Translate: read a snip’s text in another language
• Gallery and search: search, filters and sorting
• Installation: runtime dependencies, including Tesseract on Linux
• Privacy and security: what stays local and what goes online