Please see https://github.com/Shreeshrii/tesstrain-ckb It uses a modified training text based on what you sent and earlier text that I had from Pewan and other corpora.
Currently the training data includes * AWN 0-9 * AEN - ARabic numbers * No Persian numbers since some shapes are similar to Arabic Numbers Fonts do not include those which convert 0-9 to either Arabic or Persian numbers. The replace layer training is still ongoing. The eval results look much better than the official ara or script/Arabic, however I do not have any real world images for testing. ArialArial BoldTahomaTahoma Bold tessdata_fast/ara Accuracy 62.74 63.49 61.56 61.71 tessdata_fast/ara Basic Arabic 95.68 95.22 95.76 94.10 tessdata_fast/ara Arabic Extended 0.31 1.13 0.41 1.32 tessdata_fast/script/Arabic Accuracy 80.99 80.83 83.02 77.17 tessdata_fast/script/Arabic Basic Arabic 96.68 96.34 96.05 93.87 tessdata_fast/script/Arabic Arabic Extended 57.20 58.23 63.76 54.72 ckbLayer_1.661_152089_296500 ckbLayer_fast Accuracy 98.20 97.78 98.06 96.13 ckbLayer_fast Basic Arabic 99.10 99.15 98.54 98.44 ckbLayer_fast Arabic Extended 98.30 98.70 99.10 96.27 On Mon, Jan 13, 2020 at 7:17 PM Ayub Rauf wrote: > Hi, > I attached full training text with forbidden_characters in it. > really both of number types will be used and I see two type numbers > written in books but Kurdish institute verified that Arabic numbers will be > used from now on. Persian numbers written by Iranian Kurds and Arabic > number used by Iraqi Kurds but as I said numbers in ckb should be > written by Arabic type, but we have to recognize two type in OCR. > just like two types of "ك" and "ک" that written in books but now we only > use "ک". > I think these similarities won't into problem after that we can correct > letters in a spell checker. > As I said before Arial and Tahoma fonts are the most used fonts books > written by. > > > -- 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/CAG2NduWm%3DXQaxBergf5-OUE-C8jB3u12dSOPUPchRZT4w21Z-g%40mail.gmail.com.
