Computer vision ocr issue

Question

Computer vision ocr issue

Himanshu K 1

Here OCR not getting right extract text for MED EXP (Any one person) has $5,000. Result showing me $5,00 only. Same other text getting right.

Can you please resolve this issue as same type of multiple issues facing.

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2022-10-17T13:35:42.883+00:00

@Himanshu K It looks like the OCR and the READ API reads this particular field as 5,00 instead of $5,000. I am not sure what could be causing this, but you could report this form and others to the service team through an Azure support case. We will also check internally to understand more about this behavior.

I also noticed that the same form using the form recognizer works correctly and identifies the fields with the prebuilt general document API. If you are planning to extract the required fields from a document then you could consider FR API too.

1 answer

Your answer

romungi-MSFT 48,906 Reputation points Microsoft Employee Moderator

2022-10-17T13:35:42.883+00:00

@Himanshu K It looks like the OCR and the READ API reads this particular field as 5,00 instead of $5,000. I am not sure what could be causing this, but you could report this form and others to the service team through an Azure support case. We will also check internally to understand more about this behavior.

I also noticed that the same form using the form recognizer works correctly and identifies the fields with the prebuilt general document API. If you are planning to extract the required fields from a document then you could consider FR API too.

Answer 1

@romungi-MSFT Yes, that knows but my project build on Microsoft Computer Vision API and layout model form recognize have not full control to customize. Also form recognize not worked correctly as per small variations in same form.

see in image that amounts are combined.

In my original attachment of issue, there are similar other digits but why this specific $5,000 not read properly. There are multiple documents (over 500) I checked and read correct but 3-5 document having same kind of issue which degrade accuracy level.

As you mentioned, need to report this issue to service team through Azure support, so kindly provide steps about how can I post to service team about this issue?

Thank you.

Share via

Computer vision ocr issue

1 answer

Your answer