Creating a good abc file for OCR

[Mahael]

Hello,

I am attempting to programatically generate training data for my text recognition VI. I found a pre-exisiting VI (https://decibel.ni.com/content/docs/DOC-26398) which takes a string of characters and generates an .abc. However the results weren't too precise. A minor increase in detection rate was achieved by adding duplicate characters but in different size. However, "IMAQ OCR Train" is capable of handling 255 characters, so I cannot create an enormous amount of characters for the training data nor am I confident the result would be that much better for it either. I was capable of creating another .abc file by manually identifying the characters in "Vision Assistant" VI, but the difference between the methods was neglible and the latter was much more time consuming and less reusable.

My questions are the following:

How could could I create a better training data set utilizing the attached VI?

Would it be feasible to generate multiple abc files and aggregate them into one?

[udka · Accepted solution]

I think you can merge character set files in NI OCR Training interface.

[Mahael]

Thank you for the tip regarding the merging, I will generate a number of abcs and merge, hopefully it will give more precise estimates.

EDIT: Initial tests indicate that the merging significantly improved detection

[nekajcasa]

Hi, is there an option to merge two abc files inside a VI?

[Mahael]

Starting a new thread is more likely to net you an answer than resurrecting a 2+ year old thread.

I do not currently have access to vision development module so I cannot help you out, sorry.