Abin Antony – Technical Articles

Manually extracting data from handwritten notes and scanned documents has, for no little time, been a very tedious process.  Depending on stagnant OCR tools for extracting various styles of handwritten notes or from an old , smudged document were not solving the problem to its fullest.  But with the  cloud-based AI -Machine Learning solutions being rolled out by Azure, AWS, and GCP, these tasks has turned into a piece of cake!

For that matter, Textract from AWS, Azure AI Document Intelligence,  and Document AI provided by GCP do the facilitation of conversion of scanned images and handwritten content into structured data. The modern solutions employ advanced Machine Learning techniques to extract text, tables, and key-value pairs from even the most intricate of documents.

From handwritten notes to invoice processing and the analysis of old scanned archives, it’s not difficult to see how these tools provide scalable and accurate document extraction processes. Together, they save time for the industries and enable the unlocking of information hidden within their data, all done while enhancing smarter workflows and efficient operations.

Here I am  doing a comparative analysis of a few AI – ML Services for document extraction.

Extraction using AWS Textract. Approaches, Options, Advantages and Disadvantages.

Key points:

• Has separate API/SDK Functions for Synchronous and Asynchronous operations.
• Has good amount of pre-built / pre-trained models to extract content from various kinds of unstructured digital documents – Handwritten notes, Scanned Documents, Forms, Passports.  
• Need to use Asynchronous functions to extract content from multipage PDF Documents.
• As of today, for using AWS  Asynchronous Textract Functions, the  documents need to be part of a AWS S3 bucket. So, kindly make sure your architecture covers  moving documents to AWS since Async Textract operations wont accept remote or external document URL endpoints, OR another approach is to shred the document into smaller multiple documents less 5MB so that we can use byte array with Synchronous Textract functions.
• QR Codes , Bar code recognition on a document is under feature request with AWS Textract. But can use   external libraries like Python Zbar,Dotnet’s IRONQR as a additional task(with AWS Lambda), which can make workflow slower.