Thank you for replying, that was very helpful. I've now tried tessdata_best and tessdata_fast trained data found on the tesseract github, which has drastically improved my results, but still not as accurate as yours. Here are my outputs:
tesseract listpng output2 --psm 6 --tessdata-dir ~/tessdata/tessdata_best --oem 1 3 70 2 127 4 15 7 96 7 98 9 B58 9 65 19 695 29 91 33 75 tesseract listpng output_fast --psm 6 --tessdata-dir ~/tessdata/tessdata_fast --oem 1 3 70 2 127 4 15 7 56 7 58 9 #58 9 #65 19 ~=665 24 #691 33 #675 On Saturday, August 31, 2019 at 11:24:23 AM UTC-5, Jack wrote: > > I have a weird niche project here, essentially I have about 4,000 images, > each with 2 numbers between 0 and 127. > I've tweaked the images in a million different ways and I can't get > tesseract to recognized individual numbers, with the exception of 2, all > other 1 digit numbers are not recognized. > > Also, for some reason if I use tesseract directly I get way worse results, > whereas if I convert to pdf first and use ocrmypdf, which apparently uses > tesseract, I get WAY better results, which I don't understand. > > The font is very straight-forward I think, so I'm not sure if training > would be helpful, but I'm open to the idea if needed. > > Here are the sample images I'm using for testing, before and after I > modified them: > Before: https://imgur.com/a/PhjWXXK > After: https://imgur.com/a/sCRE67S > Okay some of them failed to upload but that's the gist. > > Thanks, > Jack > -- 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 view this discussion on the web visit https://groups.google.com/d/msgid/tesseract-ocr/934d89f8-a455-4787-8d8d-8986cc615059%40googlegroups.com.
