-1

I have an use case where i need to parse an image or PDF to read a survey form with all the values like Name, Age, Address etc as a key value pair and load the data to the table for each of the column.

Right now we are using AWS Textract and we are getting all the info as expected but the key, value pair for the checkboxes and radio buttons are not getting properly.

For example, the question is Are you from India? and two radio buttons yes, no.

The expected output if yes is checked is { 'Are you from India?': 'Yes' }. but the output i am getting is {'YES': 'SELECTED', 'NO': 'NOT_SELECTED'}.

Is there any way to train Textract to achieve this information or any other approach i can try to achieve.

Can someone please help me with this. I am new to this domain so i am completely blind on where to look for this.

Thanks In Advance.

Anish Raju
  • 24
  • 3

1 Answers1

0

To answer your question, no you cannot train textract like you can with comprehend or rekognition. What you need to do is figure out how to map the json returned by textract. This will differ based on the document you are analyzing. Textract returns the x and y coordinates where it detected the entity, perhaps you can use that to narrow down which category it belongs to.

Ninad Gaikwad
  • 4,272
  • 2
  • 13
  • 23
  • 1
    Thanks a lot Ninad, Sorry i just saw your response. And i did exactly the same logic to achieve the response we needed and that was enough for now. – Anish Raju Mar 03 '22 at 10:53