SikuliX

OCR issues

Asked by Jakub

I need to use OCR. Some tests I've made lately show that Sikuli' OCR doesnt work properly. Even if it comes to recognize images preprocessed with only text in it. I have some questions related with OCR engine:

- is there a possibility to train OCR engine with sikuli ( I guess a training set is a reason why it doesn recognize well)
- is there any info about this how OCR engine was trained? what fonts, styles, font sizes were used ...
- please give some info about OCR this if it really works as bad as I can see or maybe I am doing something wrong ...

I am checking text in an image with a use of TextRecognizer class and sending to recognizer a BufferedImage. Also check images generated by sikuli ide..

Best

ps. sikuli ide is just making a screen shot of the screen and cuts from it marked region - am I right? there isnt any processing here ...

Question information

Language:
English Edit question
Status:
Answered
For:
SikuliX Edit question
Assignee:
No assignee Edit question
Last query:
Last reply:
Revision history for this message
RaiMan (raimund-hocke) said : 		#1

all known problems - see linked bug

Revision history for this message
Jeff Sant (sant-jeff) said : 		#2

Can you please send me text recognizer class and the code, as i need to fix it for my issue.

Thank you

Revision history for this message
Sebastien Pinel (pinel-sebastien) said : 		#3

For the java class:
https://github.com/arnaudgelas/sikuli/tree/master/sikuli-script/src/main/java/org/sikuli/script

For the cpp that link sikuli to Tesseract OCR:
https://github.com/arnaudgelas/sikuli/tree/master/sikuli-script/src/main/native with tessocr.cpp / tessocr.h

Revision history for this message
Jeff Sant (sant-jeff) said : 		#4

Thanks Sebastien,

I have found the code of textrecognizer, do you have any working sikuli example of how to use it as i have no clue how to use it.

Thank you

Revision history for this message
RaiMan (raimund-hocke) said : 		#5

There are no examples based on the internal usage of the OCR feature. You have to dig yourself through the code (not so easy I admit ;-)

The screenshot is processed to some grayscale using openCV features, before giving it to tesseract.

Revision history for this message
RaiMan (raimund-hocke) said : 		#6

BTW: as far as I know: the OCR feature will be revised with rc3.

Can you help with this problem?

Provide an answer of your own, or ask Jakub for more information if necessary.

To post a message you must log in.