Have you tried with the existing english traineddata? I get good recognition with your 'prepared-image'?
If that is the kind of image you need to OCR, you could do that with psm 6 and then split each letter separately? ShreeDevi ____________________________________________________________ भजन - कीर्तन - आरती @ http://bhajans.ramparivar.com On Fri, Nov 14, 2014 at 7:12 PM, Simon Støvring <[email protected]> wrote: > Hello, > > I am trying to recognize single characters written with the Gotham Bold > font. I have trained Tesseract by following Michael Jay Lissners guide > "Adding New Fonts to Tesseract 3 OCR Engine" > <http://michaeljaylissner.com/posts/2012/02/11/adding-new-fonts-to-tesseract-3-ocr-engine/>. > I trained it using a newspaper article and removed all characters that I am > not interested in as well as making sure all characters are upper case as I > am not going to match lower case characters. > > I run Tesseract with my custom language and with page segmentation set to > 10, which treat the image as a single character. > > While most of the matches are fine, I am getting a lot of incorrect > matches. For example, the below image of the letter "B" is matched as an > "X". I cannot figure out why this is. > > > <https://lh4.googleusercontent.com/-AOLPnD7nXJY/VGYC58I-roI/AAAAAAAAASQ/kTJq9eSNMy4/s1600/0-4.png> > > And the "B" below which looks the same as the above but it is in fact not > the same image, is not matched to anything. Tesseract does not know what is > on the image. > > > <https://lh4.googleusercontent.com/-b0kMaAzcN-Y/VGYFI6NOzjI/AAAAAAAAASk/c9EfpR8CjWI/s1600/1-7.png.png> > > > The below "C" is not matched to anything. Tesseract cannot figure out what > is on the image. > > > <https://lh5.googleusercontent.com/-ZKl8jE2Orto/VGYEs2xzGlI/AAAAAAAAASc/2xTXomhIkWI/s1600/0-8.png> > The same goes for the "U" below. > > > <https://lh5.googleusercontent.com/-fciIyBe9bDw/VGYFRh3YBNI/AAAAAAAAASs/29WZQUHqPmE/s1600/1-8.png> > And it thinks the "E" below is a "K". > > > <https://lh4.googleusercontent.com/-ZZFkr77drgM/VGYFcDydDXI/AAAAAAAAAS0/RQ1UO8U3rOY/s1600/1-9.png> > > The above errors are just examples. There are others but I think those > four examples illustrate the quirks I'm currently dealing with. > > I manually slice the image below into images of single characters like the > ones above. Maybe a completely different approach is better? > > > <https://lh4.googleusercontent.com/-TfwZnXosqB0/VGYFjLppJ9I/AAAAAAAAAS8/Oun76IHLwks/s1600/prepared_image.png> > Does anyone know how I can improve the recognition of single characters? > I'ld like the above examples to match correctly but generally it's just not > good enough and I'ld like to know if there's any way I can improve it. > Should I train differently? Should I pass other configurations or should I > process the images before trying to recognize the characters? > > Best regards, > Simon B. Støvring > > -- > You received this message because you are subscribed to the Google Groups > "tesseract-ocr" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to [email protected]. > To post to this group, send email to [email protected]. > Visit this group at http://groups.google.com/group/tesseract-ocr. > To view this discussion on the web visit > https://groups.google.com/d/msgid/tesseract-ocr/e905020c-f0b2-47b6-b09c-e01efa96dcc1%40googlegroups.com > <https://groups.google.com/d/msgid/tesseract-ocr/e905020c-f0b2-47b6-b09c-e01efa96dcc1%40googlegroups.com?utm_medium=email&utm_source=footer> > . > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at http://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/CAG2NduWBmmBVaZp879yXz-%3D2i2ctp0MtTLciX893t-_cPRLODQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
