Restrictions for the current version of text recognition
Written by
Updated at December 14, 2023
Text recognition currently works with a number of restrictions. Make sure to consider them before getting started:
- The recognition accuracy (
confidence
) value is only calculated for aline
. For words and languages, it will be borrowed from the line's value. - You will see a single language for all recognized words, even though they are recognized in all languages from the supported model. For example, if you specify
["en", "ar"]
and the text has mostly English words, all words will be marked asen
, including those in Arabic. - The text recognition feature may have difficulties with:
- Artistic fonts.
- Vertical text (when a word is written top to bottom).
- Forms where each character is typed in a separate cell.
- Very large font size, e.g., when one word takes up half of the image.