sikuli-driver team mailing list archive

[RaiMan <question232900@xxxxxxxxxxxxxxxxxxxxx>]

Question #232900 on Sikuli changed:
https://answers.launchpad.net/sikuli/+question/232900

RaiMan posted a new comment:
at comment #6:
I got your mail and made my tests on your shot.
I did not really have problems to read the texts on the shot, but the current version has problems with number only text like in your right column. And these kind of "one-pixel" fonts are a real challenge for Tesseract principally.

This is the most important thing with Sikuli's Region.text():
the image used for OCR should only contain one line of text without any surrounding pixel clutter.
Your image contains the upper border of the text field.

To simply make some OCR tests on the screen, I use this one-liner in a separate tab in the IDE:
print "]" + selectRegion().text() + "["

this will allow you to interactively select some text on the screen and instantly see the result.
in a script, you have to add some additional adjustments for the region (have a look at the function Region.grow() in Java), to avoid pixel clutter.

If the OCR quality is not sufficient for you, then you do not have a chance to improve it currently.
If you find some rules for the bads, then you might try to compensate them by scripting.

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