0

I have a script that looks for numbers in a video by extracting 1 frame per second and process each images. I am running into some problems as with some videos, the font seems to be easily recognized by EasyOCR and some not at all. I have applied BGR2GRAY and GaussianBlur yet none of the filters seem to fix that specific font.

I have not found any info online for this so here is my question. Is there a way to "initiate" a font to EasyOCR without training the model?

c0nfluks
  • 51
  • 8
  • it's called "training". first, consult the documentation of that library. then, talk to the author of that library. – Christoph Rackwitz Aug 03 '22 at 12:06
  • I did "consult the documentation"... there's nothing to this effect, hence why I'm asking a question here. – c0nfluks Aug 03 '22 at 17:24
  • "initiating" isn't a thing. what you want _is_ training. just to be clear on that. nothing less will do, _if_ you need the network to actually be familiar with your font. -- the github mentions train/training/trainer multiple times: https://github.com/JaidedAI/EasyOCR it's even got instructions: https://github.com/JaidedAI/EasyOCR/blob/master/custom_model.md yes, it's assuming that you know some things or can learn about them -- beyond that, you should show some data that gives you trouble... you know, [mre]-kinda stuff – Christoph Rackwitz Aug 03 '22 at 17:31

0 Answers0