Workshops Use Cases Architecture Projects Blog
About Contact Offers Talks
Home›AI technologies›Reading documents (OCR)Deutsch

Reading documents: text, tables and forms out of a scan

Optical character recognition (OCR) turns a scan, a photo of a page or a PDF into text. Document services also return the layout: which word sits in which table cell and form field.

Amazon TextractAzure AI Document IntelligenceDocument AIT
Book a free 30-min call →AI Concept and Prototype workshops →

What is OCR, and what is document understanding?

Plain OCR finds the characters on a page. Document understanding services go further: they keep the structure, so a table comes back as a table and a form comes back as field names with values. Handwriting is supported by the larger services.

When the document is already structured, such as a CSV or a clean export, it needs a parser, not a model.

Goes inA scan, a photo of a page, or a PDF
Comes outThe text, plus the layout: tables, form fields, key-value pairs
Typical usesInvoices, contracts, delivery notes, claims, archives

Four ways to run it

Each of the big clouds offers it as a managed service, and you can also run it yourself. Same task, a different bill and a different console.

Amazon Textract
Managed on AWS

Amazon Textract[1]

The managed document text and layout extraction service on AWS. Single pages go through a synchronous call; multipage documents use the asynchronous API.

Official documentation →
Azure AI Document Intelligence
Managed on Microsoft Azure

Azure AI Document Intelligence[2]

The managed document extraction service on Microsoft Azure, with prebuilt models for invoices, receipts and IDs.

Official documentation →
Document AI
Managed on Google Cloud

Document AI[3]

The managed document processing service on Google Cloud.

Official documentation →
T
Run it yourself

Tesseract[4]

The long-standing open-source OCR engine, Apache 2.0. It reads the text; the layout logic around it is yours to build.

Project page →
Amazon Textract: The managed service on one page, with the same service on the other platforms.
The managed service on one page, with the same service on the other platforms.[5]

How it looks in an architecture

Slides from my course on AI architectures at FH Burgenland, drawn from real projects.

Invoices go into S3, a small function calls Textract, and the numbers land in your database. Nothing runs in between.
Invoices go into S3, a small function calls Textract, and the numbers land in your database. Nothing runs in between.[6]
The same task four ways: a ready-made service, a large language model, your own trained model, or your own platform.
The same task four ways: a ready-made service, a large language model, your own trained model, or your own platform.[6]
The same invoice into a ready-made service and into a language model: one right answer versus as many answers as prompts.
The same invoice into a ready-made service and into a language model: one right answer versus as many answers as prompts.[6]

Where I used it

Document reading is not part of my three reference projects. The invoice architecture above is a worked example from my course, so you can see the pattern before you build it.

Managed and self-hosted together: hybrid

With a ready-made service the call leaves your network. The hybrid decision is what crosses, and in what state: the whole file, a redacted one, or only the part that has to. Where a file may not leave at all, the same step runs with the open-source option inside your building, and only the result, or nothing, crosses.[7]

Hybrid is a decision at every layer. For a ready-made service the question is what crosses, and in what state.
Hybrid is a decision at every layer. For a ready-made service the question is what crosses, and in what state.[8]

What would help you next?

Learn to decide

The course “From Managed to Hybrid” teaches where AI should run: what each layer costs, what you operate and where your data has to stay. It is in preparation, with a free live session first.

Join the course waitlist →
Linda MohamedLinda Mohamed
AI and cloud consultant, AWS Community Hero, Vienna

Build it with me

Tell me what you want to build. In 30 minutes we check which of the four ways fits your data, your rules and your budget, and whether AWS funding for a proof of concept can cover part of it. How AWS funding works →

Book a free 30-min call →15-min question slot

You already know what you need

Book the workshop directly:

ai-solutions.wiki

Read more on ai-solutions.wiki

My open knowledge base, with longer comparisons, pricing notes and sources.

Sources

  1. Amazon Textract documentation. https://docs.aws.amazon.com/textract/latest/dg/what-is.html
  2. Azure AI Document Intelligence documentation. https://learn.microsoft.com/en-us/azure/ai-services/document-intelligence/overview
  3. Document AI documentation. https://cloud.google.com/document-ai/docs
  4. Tesseract. https://github.com/tesseract-ocr/tesseract
  5. Linda Mohamed, course slides “The projects”, MASE-3, FH Burgenland, September 2026
  6. Linda Mohamed, course slides “How to run AI, layer by layer”, MASE-3, FH Burgenland, September 2026
  7. NIST glossary: hybrid cloud. https://csrc.nist.gov/glossary/term/hybrid_cloud
  8. Linda Mohamed, course slides “Start here”, MASE-3, FH Burgenland, September 2026
  9. Amazon Textract, ai-solutions.wiki. https://ai-solutions.wiki/tools/amazon-textract/
  10. Azure AI Document Intelligence, ai-solutions.wiki. https://ai-solutions.wiki/tools/azure-form-recognizer/
  11. Textract vs Comprehend, ai-solutions.wiki. https://ai-solutions.wiki/comparisons/textract-vs-comprehend/
  12. Pre-built AI services vs foundation models, ai-solutions.wiki. https://ai-solutions.wiki/comparisons/prebuilt-ai-services-vs-foundation-models/
  13. Amazon Textract asynchronous operations, AWS documentation. https://docs.aws.amazon.com/textract/latest/dg/async.html

Links checked September 2026. Product names and features change: check the linked pages before you decide.