Simple repetitive sequence but inaccurate OCR

Hi Guys,

I’m Alessandro and I’m working on some automation of my locomotives’ maintenance platform.
I’m working on a Citrix virtual desktop where I can’t get full text, so I’m trying to get the mileage value of each locomotive. Those are 72 numbers but each one is in a different window.

I built a simple but long sequence to get those value, and while the clicks and the operations run smoothly, the OCR is so bad.
I tried changing scales and profiles, either with MicrosoftOCR and Tesseract, but I get strange numbers and symbols. This is so bad because I see the numbers are in boxes and don’t look difficult to be read.

I’m attaching the screenshot of on procedure. After each step I send the command “arrow down” and UIPATH read the mileage of the next locomotive and so on.

Please let me know if there is a 100% accurate way to achieve my goal, thanks a lot!!


OCR is fairly unreliable for “any character” inputs. However, if you’re looking specifically at numbers, you can set the “allowed characters” to numbers in Tesseract OCR.

It is still, in my experience, fairly unreliable (5 and 6 tend to look very similar to OCR), but it’s usually a better overall result.

1 Like

Yes, I’m almost there now!
I’ve done it as a Native Citrix > Get OCR Text (Tesseract) and I’m allowing the characters -0123456789.

I see the biggest similarity is betwen 3,8,9 but only in certain position, like when they are in the 3rd and 4th to last position, so they result even if inaccurate could be acceptable.

Old post but good update.

Google vision is perfect for citrix OCR, 100% success rate!

No OCR gives you 100% Successrate