0

A PDF Document like Invoices (Name of the Product, number of cartons, total price, Unit price, etc.), Delivery Order (Name of the goods with Quantity) and etc.

I tried many methods with Python like reading the PDF file and verifying the details, but the problem is it's not accurate and it works sometimes and sometimes it does not.

What I am expecting is, a program which reads the PDF Doc very accurately and validates the document.

double-beep
  • 5,031
  • 17
  • 33
  • 41
Tushar
  • 33
  • 5
  • @KJ, it's simple. PDF like Invoices with table with Name of the product, Quantity, Total Price and etc. I tried using some APIs, it does print everything, but it's not organized. The organisation of data is random sometimes so its difficult to extract. – Tushar Jan 04 '23 at 19:05
  • Can you suggest any solution for the problem described? – Tushar Jan 04 '23 at 19:43
  • 1
    You can use DocumentAI which is designed for that. and 100% is not reachable ;) – guillaume blaquiere Jan 05 '23 at 05:11
  • @guillaumeblaquiere, how's AWS Textract? – Tushar Feb 08 '23 at 10:07
  • I absolutely don't know! I'm a Google Cloud expert and I rarely worked with AWS, at least not enough to answer you (to be honest, I just discover the AWS service name with your comment...) – guillaume blaquiere Feb 08 '23 at 10:09
  • You know when I went to Document AI page and there was a Try It! feature, so I tried using it an it was working. But the problem, there are various ways to extract data for e.g by choosing on invoice, Document OCR, Form Parse, Table but there was some data which was not getting extracted in each of these and with Document OCR the result was unstructured. This is where AWS Extract excels. Any advice? – Tushar Feb 08 '23 at 13:46
  • FYI, Document AI has an actively monitored tag [`[cloud-document-ai]`](https://stackoverflow.com/questions/tagged/cloud-document-ai) --- – Holt Skinner Mar 28 '23 at 22:14

1 Answers1

0

I recommend looking into the Invoice Parser for Document AI, which is a model specifically designed for handling this format.

Refer to Send a processing request and Handling the processing response for information on using the API.

If you want 100% accuracy, it's not going to be possible to get that consistently from any OCR solution. If you want to increase accuracy of the pretrained processors, you can use Uptraining to add your own training data and make custom models based on the existing models.

You can also utilize Human-in-the-Loop for production review to get human verification when you have high-accuracy requirements.

double-beep
  • 5,031
  • 17
  • 33
  • 41
Holt Skinner
  • 1,692
  • 1
  • 8
  • 21