How to improve accuracy of OCR?

Anurag Singh 0

Hello,

is there any way to label and train data in the custom model without manually updating ocr.json and label.json .

I'm currently facing issue with these characters,

Character having high failure frequency: -

1 (as I, J and T),

5 (as S),

D (as 0 and O)

Character having lower failure frequency: -

F (as E),

L (as 1),

Q (as O),

T (as 1 and 7),

Z (as 2)

is there any way to fix atleast character having high failure frequency?

Thanks

VasaviLankipalle-MSFT 17,471 Reputation points

2024-05-07T00:16:56.64+00:00

Hello @Singh, Anurag {PEP} , Thanks for using Microsoft Q&A Platform.

Is it possible to share the model and the API version you are working with?
Anurag Singh 0 Reputation points

2024-05-07T05:53:49.48+00:00

Thanks for replying @VasaviLankipalle-MSFT

first, I used prebuilt-read in latest GA and preview version both.

Then I created custom model also with same GA (2023-07-31) and preview (2024-02-29-preview) version.

Results are same for all.
Anurag Singh 0 Reputation points

2024-05-08T05:51:56.0433333+00:00

Hi @VasaviLankipalle-MSFT , any update?

Your answer