sikuli-driver team mailing list archive

[RaiMan <question247086@xxxxxxxxxxxxxxxxxxxxx>]

Question #247086 on Sikuli changed:
https://answers.launchpad.net/sikuli/+question/247086

    Status: Open => Answered

RaiMan proposed the following answer:
There are many quirks with the current implementation of the OCR feature (known problems, some bugs reported)
It is based on Tesseract, that has a training feature, but this currently is not available with Sikuli in any way.
You would have to do it on your own and integrate the train-data into the libs/tessdata folder.

Another option is to try to correct the resulting strings afterwords, based on known failure patterns.
In your case: If you know the resulting string should be a number and the 1 is always read as lowercase I, then you could correct this afterwards.

some things, that might improve the results:
- only one line of text
- dark text on light background (best: black on white)
- font size >9 points
- regular/common fonts
- the region to be read should contain as little background as possible and nothing else than text

-- 
You received this question notification because you are a member of
Sikuli Drivers, which is an answer contact for Sikuli.