X-1.0rc2: Text extraction from Images fails in some cases on colored backgrounds

[praveen]

I need to fetch numbers from different labels having different colors for the purpose of validation.
While doing so, Sikuli extracts numbers only from few labels(having similiar color shades).
So,I believe extraction got failed due to the color shades behind the numbers.
Kindly provide a fix/ add this feature in upcoming release.

[Alan (alan-ciusky)]

I've had experienced the same issue.

Any variance on the background (e.g: Gradient) will cause the OCR to fail.

I've been trying to workaround this issue by manipulating the image to eliminate background.

[MadhusudanG (g-madhusudan)]

I am also facing the issue but there is no variance in background. It is just black and white space chars.

On the screen, the text is:
What city were you born in?

But when extracted the text from the screen, the text is:
what my were you born nn?

My code to extract questions: //png contains the snapshot of all questions in the page
--------------------------------------------
Iterator<Match> match = s.findAll("D:/Cygnets_CVS/qa/QAIndia/xyz_Automation/ft/src/main/java/com/arcot/test/mobile/android/imgs/QnAPage2.png");
Thread.sleep(1000);
qlist=match.next().text();
System.out.println(l + ": " + qlist);

[Gregory Peck (gmsf)]

I had similar problems. I tried to extract numbers from regions on screen. The numbers were not correctly identified using region.text() and it seems if numbers are white on dark background it fails completely and often the numbers where identified as different text.