Extract Text with OCR - Use another language to specify "trained" data

[cjibb02]

Hi,

I'm using the "Extract text with OCR" action to read text from a grid in an application running under Citrix. I'm getting OK-ish results after following the suggestions in this post - https://powerusers.microsoft.com/t5/Power-Automate-Desktop/Move-mouse-to-text-on-Screen-OCR-worked-y...but still not good enough.

The text that i need to read (and then compare to an input variable so it needs to be exact) aren't real words in any language.  It also has lots of slashes e.g. C560XL/XLS/IR. I was thinking that if i could train the engine with this text then i would get better results. Power Automate Desktop has the "Use Other Language" option and allows you to set the language data path. I can't find any instructions on how this works and i keep getting "Failed to create the OCR engine" error.

Can anyone give me more information on this feature? Will it do what i want it to do? How do i specify the data file?

I've tried downloading the language data from here - https://github.com/tesseract-ocr/langdata but i'm clearly doing something wrong ...

Any help will be much appreciated!

[fraenK]

For the additional language did you try this? https://docs.microsoft.com/en-us/power-automate/desktop-flows/how-to/ocr-multilingual-documents

BUT the built-in OCR functionality based on Tesseract is not that great.

Did you try the Microsoft cognitive action for OCR?

https://docs.microsoft.com/en-us/power-automate/desktop-flows/actions-reference/microsoftcognitive#o...

Or would there be a chance to export the grid content as text from the application itself or take a screenshot and run it through a more advanced 3rd party OCR tool install PAD within Citrix?

Unfortunately other RPA product are more advanced with Citrix based automation.

[cjibb02]

Thanks, yes I've followed the steps in that link but i always get "Failed to create OCR engine" even when i use the language files downloaded from github - https://github.com/tesseract-ocr/langdata. I have a support ticket open with Microsoft for this and will update with what they come back with. 

I haven't tried the Microsoft OCR Cognitive action yet but will take some screen shots and run these through that service. There will be challenges identifying and cropping the part of the screen i want to run the OCR on, saving as an image and then parsing the results. So I'm not sure if it will be practical but will be interesting how it compares to Tesseract over the whole screen image.