OCR: improve recognition accuracy and word count

(0) Share

Report

Posted on by BLabat83

122

Hello to all,

As always, first of all thank you very much for your help. Today I am here to ask for help with OCR:

1. I need to position the mouse on a text, but the OCR fails a lot, even when it is text on the screen. any ideas to improve its recognition accuracy?

2. Is there any way to count the number of times a text appears on the screen? I understand that I can use the OCR function to get the text, but how do I then count the words I want to search for?

Categories:

Power Automate Desktop

I have the same question (0)

All responses (8)

Answers (1)

Sort by

Verified answer

Agnius Bartninkas Most Valuable Professional on at

Like
a
(1)

Report
Copy link

Link copied!

OCR is in general not a very accurate way of doing things and PAD is in general not great at OCR. There's not much you can do about its accuracy. You could try playing around with different text width and height modifiers inside the flow, until you get it to work (i.e. having several instances of the same action until it works), but in general it will never be very accurate.

What is it exactly that you are trying to do though? Perhaps it is possible to do the same in a better way?

As for counting the number of times some text is visible on the screen, you could extract the text with OCR and then use Parse text on it with the "First occurrence only" toggle disabled:

This will return a list of positions of the text you were looking for. You can use %Positions.Count% to get the number of matches.
-------------------------------------------------------------------------
If I have answered your question, please mark it as the preferred solution. If you like my response, please give it a Thumbs Up.
I also provide paid consultancy and development services using Power Automate. If you're interested, DM me and we can discuss it.

Was this reply helpful? Yes No
BLabat83 122 on at

Like
a
(0)

Report
Copy link

Link copied!

Thank you very much as always @Agnius .

On a screen that opens, I have to choose a line, and I had thought of using the person's date of birth, but when I use the OCR, it doesn't read the numbers properly.

So, I have thought that I can choose based on the name of the person, but, as there may be people with the same name, my proposal is that if the OCR finds the name several times, it will give an error so that it can be managed manually.

That is why I wanted to know if it is possible to improve the accuracy of the OCR of the PAD.

Any other idea?

Was this reply helpful? Yes No
Agnius Bartninkas Most Valuable Professional on at

Like
a
(0)

Report
Copy link

Link copied!

What sort of a screen is that? Is it a local app or a web page? I think there should be better options to use instead of OCR, such as extracting the values, finding the right value and then clicking on the required element.

Was this reply helpful? Yes No
MichaelAnnis 5,727 Moderator on at

Like
a
(0)

Report
Copy link

Link copied!

I would keep your conversation going with @Agnius as the search for a different solution than OCR is probably the best way to go.

if you get stuck with OCR, then I used to improve OCR accuracy by using the “maximize” features in the advanced settings. I specifically remember this being an advanced setting under the Tesseract Engine.

basically, it zoomed in the text. And the larger I made it, 2 things happened. 1) it got much more accurate, but 2) it got very slow, so just keep that in mind if you use that option.

Was this reply helpful? Yes No
BLabat83 122 on at

Like
a
(0)

Report
Copy link

Link copied!

First of all, many thanks to both of you.

@Agnius Let me tell you a bit more. It is a local application called Scandox that is used to save documents in an indexed way. As I said before, at one point in the process, a window opens to select the person to whom these documents are going to be associated (I attach a screenshot with an example with modified data). As I said, I wanted to use the OCR to use the date of birth, but this information is not recognised correctly. That's why I thought that I could use the name that it recognises it better, but in case there are several, it would give an error.
Is there any other way? thank you very much
@MichaelAnnis What is this option to maximise the text? Can you give me a screenshot? Thanks!

Was this reply helpful? Yes No
Agnius Bartninkas Most Valuable Professional on at

Like
a
(0)

Report
Copy link

Link copied!

Can you interact with the grid using the UI actions? Are there any navigation buttons available in the app that would allow you to navigate through the grid? Is there a Find functionality available, such as sending a Ctrl+F? Or a filter?

Was this reply helpful? Yes No
BLabat83 122 on at

Like
a
(1)

Report
Copy link

Link copied!

No, there is no buttons to navigate or to find or filter....so basic the window.

But I have just discovered an option, which is to get all the information from the screen, and split it up for comparison. Then with the scroll down, I can select it. It's going to be a bit of a long development but I think it's going to be worth it.
I'll let you know how I'm progressing. Thanks for your help!

Was this reply helpful? Yes No
MichaelAnnis 5,727 Moderator on at

Like
a
(1)

Report
Copy link

Link copied!

Best of luck!

Was this reply helpful? Yes No