Thank you Tom and art for you guy's kind advices. Eventually I have obtained a much better result by using a training data from openalpr project. In fact, the font in my place (hk) is different from US (HK font is more square like),then I have changed my character image's aspect ratio so it is more like that of US. I am really fortunate that this little trick work.
On Wednesday, March 30, 2016 at 11:34:14 PM UTC+8, Alex Szeto wrote: > > I am working on a license plate recognition project, I have trouble in > improve accuracy of OCR. > Attached is one of the image I used and the result is very poor. > > version of tesseract : 3.0.3 > The command that I used : tesseract Untitled.jpg out -psm 9 > The result is : SXUSBBB while I am expecting for 5X0S888 > I have did some experiments and I have found some character pairs are > easily get confused by tesseract. > for example : '0' become 'U' ; '5' and 'S' ; 'B' and '8' > > Is there some methods or parameters I can set so the result can be > improved? > Thank a lot and I really appreciated any advises. > > -- 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 https://groups.google.com/group/tesseract-ocr. To view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/c92a7d15-35ff-495f-954a-3c33667cd338%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
