I had a similar problem. You should do 2 things: 1) Remove the background - any color which is not light blue or similar to it should be removed - it is easy to implement in C++ 2) Train tesseract to recognize digits only with your particular font. Instead using "English" or "German" as a language, train your own language called "digits" that is trained on digits only. This will do the job.
Regadrs, Svetlin Nakov Managing Partner Consulting and Information Technology Agency http://www.citagency.eu -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Georg Oberth Sent: Thursday, September 17, 2009 10:46 PM To: [email protected] Subject: AW: Need to OCR a number display No actually I'm trying to read a meter display. To make things worse they are white numbers on black bkg. I am using 2.04 (which I believe is the latest version). What is svn? Do you know the variable name for "numbers only"? My numbers are 100 pixels tall - is that a problem? Sorry for all the questions. Thank you Georg Oberth [email protected] www.godata.at 0043-676-692-6070 0043-3127-88121 -----Ursprüngliche Nachricht----- Von: [email protected] [mailto:[email protected]] Im Auftrag von SteveP Gesendet: Thursday, September 17, 2009 9:13 PM An: tesseract-ocr Betreff: Re: Need to OCR a number display One other thought: You have the words "number display" in the title of this thread. Are you using something like a 7-segment LED numeric display? If so, that might require training. About 6 months ago, someone posted about that. On Sep 17, 12:03 pm, "Georg Oberth" <[email protected]> wrote: > Hi Steve, > Thanks a lot for the info. I'll try it right away. > Have a nice day! > > Georg Oberth > [email protected] > 0043-676-692-6070 > 0043-3127-88121 > -----Ursprüngliche Nachricht----- > Von: [email protected] [mailto:[email protected]] > Im Auftrag von SteveP > Gesendet: Thursday, September 17, 2009 8:57 PM > An: tesseract-ocr > Betreff: Re: Need to OCR a number display > > For numbers only I suggest getting the latest source version using > svn, > using the traineddata file that it contains instead of training > tesseract, and setting the variable for 'numbers only' using a > config. Resize your image if necessary to make the digits at least 20 > pixels high. (Even without specifying 'numbers only', I get great > results on numbers from the above approach.) > > On Sep 17, 11:35 am, georg <[email protected]> wrote:> Hi All, > > I have to OCR files that contain numbers only. > > I took a bunch of them and assemble them into a tif. > > Then I used Owlboxer to generate a box file and then proceeded to > > create all the goofy files needed as per the instructions > > inhttp://code.google.com/p/tesseract-ocr/wiki/TrainingTesseract > > > > > The thing fails miserably. > > Any suggestions? What am I doing wrong? > > Thanks a lot oin advance for your help. > > Georg Oberth- Hide quoted text - > > - Show quoted text - --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To post to this group, send email to [email protected] To unsubscribe from this group, send email to [email protected] For more options, visit this group at http://groups.google.com/group/tesseract-ocr?hl=en -~----------~----~----~----~------~----~------~--~---
