Hey,
so I am using ABBY Fine Reader 15 OCR editor and there are certain characters in my scanned document that the programme marks as low confidence. They pop up frequently, which is makes the OCR process quite time-consuming. I have been trying to fix this by using pattern training, yet whenever I am pattern training, the programme does not give me the option of training the characters that it had marked as low confidence. They just don´t pop up during training.
So is there any way of manually teaching the programme certain characters? Whenever I selectively recognize those low confidence characters, FineReader has no problem reading them.
Repeated low confidence characters don´t appear during pattern training
Was this article helpful?
0 out of 0 found this helpful
Comments
5 comments
Hello Jacob,
The issue you have described is already known to our Developers. At the moment there is no option to make FineReader check every symbol, only uncertain symbols are suggested for the training process. I have forwarded your comment to our R&D team.
To save some time I suggest you train FineReader only on a short paragraph which you can draw manually and start the training process specifically for small parts of the text (more about areas you can read in our help).
Hi, I am still seeing this issue with the latest trial version: I trained on custom language plus English, and for some reason the model never asks for input on S, 2, ? (and a few others), which subsequenty trash the result by appearing as low-confidence ^. Selecting small parts does not help, those characters are simply always skipped.
So after further investigation, the option to use Built-in patterns (in addition to mine) does not make any difference. Using built-in patterns, I get very good result on the bulk of the text, but I trained on special characters (ñāīūḍḷṁṃṅṇṭ) which are not recognized at all. When I switch to trained patterns, the result is terrible on regular text, but good on foreign words. Checking "Also use built-in patterns" makes no difference at all.
Is it a limitation of the trial version which I use? I would be happy to buy, but this is a show-stopper.
I do have the same issue. There is no way to force the training on every single character of words or lines that OCR is missing regularly even though they appear as low confidence characters after "recognize". On the other hand during training it keeps checking over and over again the same list of characters. I have trained hundred times characters like : # c r.
But some low confidence characters rarely get trained. Not even the corrections made during "verify" seem to be considered as training !
I had bought Finereader 15 for the OCR capability !
Does version 16 make a better job ?
Hello Philippe!
Thank you for your feedback! I'll forward it to the product development team.
The corrections done during verification do not train FineReader for characters, it's just editing of the recognized text of the current document. However, during verification process you can add non-dictionary words (marked in red) to the dictionary, which dictionary is used for all documents, so that FineReader doesn't stop at those words in the future.
Pattern training functionality is the same in the versions 15 and 16.
Please sign in to leave a comment.