Restrictions for the current version of text recognition
The service is currently running with a number of restrictions. Please read them before you start working:
- The recognition
confidencevalue is only calculated for a
line. The line's value is used both for words and for the recognized language.
- A single language is indicated for all recognized words, although they are recognized in all the languages from the supported model. For example, if you specify
["en", " ar"]and the text has mostly English words, then all the words are marked as
en, including those in Arabic.
- For good recognition quality, text shouldn't be too slanted. Ideally, the angle should be less than 10-15 degrees.
- The following is not recognized well:
- Handwritten text.
- Artistic fonts.
- Vertical text (when a word is written from top to bottom).
- Any set of letters and numbers, like a license plate number.
- Forms where each character is typed in a separate cell.
- Short words and numbers in table cells.
- Very large text, for example, when one word takes up half of the image.