Textract

📄 What is Amazon Textract?

Amazon Textract is a machine learning-based OCR service that automatically extracts printed and handwritten text, tables, forms, and key-value pairs from scanned documents, PDFs, and images — without needing templates or custom code.

✅ It goes beyond basic OCR by understanding the structure and relationships in your documents.

---

🎯 Key Use Cases

Industry	Use Case
Banking	Extract KYC data from ID documents
Healthcare	Digitize handwritten prescriptions or records
Insurance	Read claim forms and policy documents
HR	Process resumes, forms, or offer letters
Legal	Extract clauses from contracts
Logistics	Read invoices, receipts, and bills of lading

---

📚 What Textract Can Extract

Feature	Description
Raw Text	Plain text from PDFs, images, scans
Forms	Key-value pairs (e.g., “Name: John”)
Tables	Structured rows and columns
Handwriting	Supports both printed and cursive
Checkboxes	Detects selection states (ON/OFF)
Identity Documents	Structured fields from ID cards, licenses
Expense Analysis	Categorize receipts, totals, vendors, etc.

---

🧠 Textract API Operations

API Method	Purpose
DetectDocumentText	Extract raw lines and words
AnalyzeDocument	Extract forms, tables, checkboxes
AnalyzeExpense	Specialized for receipts, invoices
AnalyzeID	Specialized for identity documents
StartDocumentAnalysis	Async operation for large docs
GetDocumentAnalysis	Retrieve results from async jobs

---

🧪 Sample Python Code (Boto3)

Extract Form & Table Data (Sync)

import boto3

textract = boto3.client('textract')

with open("form.png", "rb") as document:
    image_bytes = document.read()

response = textract.analyze_document(
    Document={'Bytes': image_bytes},
    FeatureTypes=["FORMS", "TABLES"]
)

for block in response['Blocks']:
    if block['BlockType'] == 'KEY_VALUE_SET':
        print(block.get('Text'))

---

Extract from S3 (Async for Large Docs)

response = textract.start_document_analysis(
    DocumentLocation={'S3Object': {'Bucket': 'my-bucket', 'Name': 'invoice.pdf'}},
    FeatureTypes=["TABLES", "FORMS"]
)

job_id = response['JobId']

---

🧾 Output Format Example

Textract returns nested blocks with relationships:

{
  "Blocks": [
    {
      "BlockType": "KEY_VALUE_SET",
      "EntityTypes": ["KEY"],
      "Text": "Name"
    },
    {
      "BlockType": "KEY_VALUE_SET",
      "EntityTypes": ["VALUE"],
      "Text": "John Smith"
    }
  ]
}

---

🧮 Textract vs Other OCR

Feature	Textract	Simple OCR (e.g., Tesseract)
Text	✅ Yes	✅ Yes
Tables	✅ Yes	❌ No
Forms/Key-Value	✅ Yes	❌ No
Handwriting	✅ Yes	Limited
Identity Docs	✅ Prebuilt parser	❌ Manual
Expense Docs	✅ Yes	❌ No

---

🧾 Pricing (2024)

API	Price per Page
DetectDocumentText	$0.0015/page
AnalyzeDocument	$0.015/page (tables/forms)
AnalyzeExpense	$0.05/page
AnalyzeID	$0.025/page

🧠 Free Tier: 1,000 pages/month for first 3 months (analyze or detect text).

---

🔐 Security

Feature	Support
IAM integration	✅ Yes
Encryption	✅ With KMS
S3 Integration	✅ Yes
VPC endpoint support	✅ Yes (via PrivateLink)
HIPAA eligible	✅ Yes

---

🧱 Terraform Integration (Indirect)

No direct Terraform resource for Textract, but you can:

• Automate Textract through Lambda functions

• Trigger it via S3 events using:

resource "aws_lambda_function" "textract_handler" {
  ...
}

resource "aws_s3_bucket_notification" "notify" {
  bucket = "my-textract-bucket"

  lambda_function {
    lambda_function_arn = aws_lambda_function.textract_handler.arn
    events              = ["s3:ObjectCreated:*"]
    filter_suffix       = ".pdf"
  }
}

---

⚙️ Integration With Other Services

Service	Purpose
S3	Store source and output files
Lambda	Trigger processing logic
Comprehend	NLP on extracted text
Athena	Query structured output
Step Functions	Orchestrate large pipelines
OpenSearch	Index documents for search

---

✅ TL;DR Summary

Feature	Amazon Textract
OCR for scanned docs	✅ Yes
Forms/key-value pairs	✅ Yes
Tables	✅ Yes
Handwriting	✅ Yes
Identity/receipts	✅ Specialized APIs
Real-time + batch	✅ Both supported
Output format	JSON with structured blocks
Free Tier	✅ 1000 pages for 3 months

---