Text Recognition Prebuilt model

(0) Share

Report

Posted on by sahachiranjit32

143

Hi All,

I'm using Text recognition prebuilt model to capture some of text from Images. When I feed attached image(images 124) to prebuilt flow it gives all extracted information.

I want to send the Information (Nutritional Values) in Table format. can anyone please suggest how to do it?

Categories:

AI Builder

I have the same question (0)

All responses (5)

Answers (0)

Antrod Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Hi @sahachiranjit32 ,

Thanks for reaching out.

There is no automated way to get a table from the Text recognition model, you will just get the raw text from the document. What you can do however is to list all the lines and send the content as a list.

If you manage to split the lines of text into relevant blocks of text using the Power Automate functions, you could build an HTML table with these blocks of text using the "Create HML table" action in your flow and then send it by email for example.

Was this reply helpful? Yes No
sahachiranjit32 143 on at

Like (0)

Report
Copy link

Link copied!

Hi @Antrod ,

Thanks for the suggestion.. We have lot those Images where some Table values start with Nutrition Value/Nutrition Info/Daily Value like that.. Even If I split lines how do I create HTML Table?

I think probably I need to train the model for that...and we have 1 Image for every customer not 5 (I have attached some samples) and design is completely different from each other.. Do i need to copy paste same Image 5 times and train model for all different Image as Structured Documents ( In different collection)?

EDIT: Even If I upload same Images 5 times changing Image Name, It says Document already uploaded

Need your suggestions. Thanks for your help

Was this reply helpful? Yes No
Antrod Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

If you have different image design to consider, the Text recognition model won't help to structure correctly the data and you should definitely try the custom Document Processing model.

You need 5 different samples for each image design. Changing the file name doesn't count for a different sample (content should actually be different). You could perhaps try the unstructured version of the document processing model including only files of different design but not sure about the result. Worth a try though 🙂

Was this reply helpful? Yes No
sahachiranjit32 143 on at

Like (0)

Report
Copy link

Link copied!

Hi @Antrod ,

I have trained the unstructured model , but the accuracy score is poor.. It's not all acceptable, any other way around we can do the automation for these kind of documents?

Was this reply helpful? Yes No
Antrod Microsoft Employee on at

Like (0)

Report
Copy link

Link copied!

Hi @sahachiranjit32 ,

Unfortunately no, training the model requires several samples of same design.
The only possibilities I see are quite manual/complex:
- Do complex string manipulation after using Text recognition to extract nutritional value for each possible sample design.
- Generate manually multiple samples of same design by editing the image with an image editing tool so you can use them for training a document processing model

Was this reply helpful? Yes No