This is a known issue with Tesseract. One solution is to process the OCR results then detect the size discrepancy between the two parts of the line and then re-process each part as a separate image. In essence, doing that prevents Tesseract from drawing bad inferences.
I think Tesseract 3.01 brings improvements to this issue from initial tests I have run. I will try your images and post back here with my findings. Patrick On Mar 9, 9:56 am, Søren Engel <soren.en...@gmail.com> wrote: > Hello fellow members, > > I am currently working on a upgrading an old OCR module at our > development team which was originally written in VB using the > integreted OCR components within Microsoft Office 2003. Since this is > discontinued (as Microsoft was discarded the COM components from the > office distributions), I have been trying to locate a fitting module > to take to spot - leading me to this site and binaries. > > My problem now stands as follows, while performing a OCR on a basic > invoice coming from a fax using the tesseract-ocr commandline app, > whereas the invoice contains a sentence like > > -- contract number: 214587 -- > > I noticed, since the numbers may in some cases by twice the size as > the other text, that the text just messy and the numbers are not > included in the output. I somehow suspect that the OCR has a upper > threadshold limit on how large a character must be, compared to other > characters in the image, before it is considered not to be text, thus > discarded from the output. > > Since I am both new to the concept of OCR as well as our team don't > have members with a strong background in image processing, I am a bit > unsure how to tackle this issue. > > The images I have been using for my initial tests may be found at > > http://www.ge.tt/4KxYgTUhttp://ge.tt/29sDgTX > > I should mention that the image test.tif is at its original size and > gives bogus results, where as the test_01.tif is scaled about 33% of > the original size and produces the correct result. However, it is not > a viable solution just to scale the images, since this may lead to > incorrect results for the rest of the document (note: not everything > is included in these images due to privacy restrictions). > > I should also mention, that the old microsoft COM component had no > problems handling these images. > > I hope that I can get some feedback on this matter. > > Cheers, > Kind regards > Søren Engel -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To post to this group, send email to tesseract-ocr@googlegroups.com. To unsubscribe from this group, send email to tesseract-ocr+unsubscr...@googlegroups.com. For more options, visit this group at http://groups.google.com/group/tesseract-ocr?hl=en.