Image

Reading the text inside a photo, a screenshot, or a scanned page. The action is called Extract text, and it works offline.

Image to text

• Recognition languages — which of the installed languages are used. More than one can be ticked, and the last one cannot be unticked, because recognition needs a model. The row appears once you have two or more languages installed.
• The installed languages are listed under it. Activating one offers to delete it.
• Download a language — the last row. Twenty languages are offered, Persian and English first.

Extract text does not appear in the message menu until a language is installed, so it is never an action that can only fail. If recognition reports a missing language it says so and offers the download.

Choosing languages well

• Persian alone reads Persian most accurately. Adding English also reads Latin text, but a Persian word is occasionally returned as Latin nonsense. This is how the recognition engine works, not a setting that can fix it — so it is your choice which trade you want.
• The larger the text is in the picture, the better it is read. A screenshot at full size reads better and faster than the same text photographed small.
• Dark backgrounds are handled for you: a dark picture is inverted before recognition, because the engine is trained on dark text over light paper.