Amazon Textract

What it is

Amazon Textract is a pretrained document analysis service. It extracts printed and handwritten text from scanned documents and images and, beyond plain OCR, understands document structure: forms (key-value pairs), tables, and specific fields. It works with financial reports, medical records, tax forms, invoices, receipts, and identity documents. No ML (Machine learning)Software that learns patterns from example data to make predictions or decisions, instead of following rules written by hand. A model is first trained on data, then used to make predictions on new data, which is called inference.Glossary entry expertise is needed.

Choose Textract when a question asks to extract text or data from documents, such as scanned PDFs, forms, invoices, or IDs, to replace manual data entry. For text in everyday photos, such as street signs, Amazon Rekognition also works. To understand the meaning of the extracted text, such as sentiment or entities, pass it to Amazon Comprehend.

APIs

API	What it extracts
DetectDocumentText	Lines and words of text
AnalyzeDocument	Text plus forms (key-value pairs), tables, signatures, and answers to Queries
Queries (a feature of AnalyzeDocument)	Ask in plain language for the field you need, such as "What is the policy number?". Custom Queries adapt the pretrained feature to your own documents
AnalyzeExpense	Fields from invoices and receipts, such as vendor, total, and line items
AnalyzeID	Fields from identity documents issued by the U.S. government, such as driver licenses and passports
Analyze Lending	Classifies the pages of mortgage loan packages and routes each page to the right analysis

Synchronous or asynchronous

Synchronous operations process single-page documents (JPEG, PNG, PDF, TIFF) with near real-time responses. They suit interactive apps.
Asynchronous operations process multipage PDF and TIFF documents stored in Amazon S3:
1. Call a Start operation, such as StartDocumentAnalysis.
2. Textract publishes the completion status to an Amazon SNS topic, which an SQS queue or a Lambda function can read.
3. Get the results with the matching Get operation.

A common serverless pattern: a document lands in S3, an S3 event invokes Lambda, Lambda starts an asynchronous Textract job, the SNS notification triggers a second step that stores the results in DynamoDB or S3, and Comprehend analyzes the text.

Security, resilience, performance, and cost

Security

IAM controls access to the API (Application programming interface)A defined way for one program to ask another for data or actions, such as a set of HTTPS requests and responses. Every AWS service is controlled through its API, which the console, the CLI, and the SDKs call for you.Glossary entry and to the S3 documents Textract reads.
Input documents in S3 are protected with S3 SSE (Server-side encryption)Encryption that a service applies when it stores your data and removes when you read it back. Amazon S3 encrypts every new object with SSE-S3 by default; SSE-KMS uses keys in AWS KMS instead.Glossary entry (S3-managed keys or KMS keys); traffic uses TLS (Transport Layer Security)The protocol that encrypts traffic between a client and a server and lets the server prove its identity with a certificate. It replaced SSL, though the older name is still common; HTTPS is HTTP over TLS.Glossary entry.
Interface VPC endpointA private connection from a VPC to an AWS service or another endpoint service without crossing the internet. Gateway endpoints, for Amazon S3 and DynamoDB, are route table targets; interface endpoints are network interfaces that use AWS PrivateLink.Glossary entry (AWS PrivateLink) keep calls off the internet. CloudTrail records API calls.

Resilience and performance

Textract is a managed, Regional service. Use the asynchronous APIs with SNS and SQS for large or bursty volumes, and retry ThrottlingLimiting how many requests a client can make in a period. A throttled request is rejected, often with HTTP status 429 or a throttling error, and should be retried later with backoff.Glossary entry requests with backoff.

Cost

Pay per page processed. The price depends on the API and features used: plain text detection costs less than forms, tables, or queries.

Monitoring

CloudWatch metrics for successful requests, errors, and throttling; CloudTrail for auditing.

Textract or another option?

If the requirement is	Choose
Extract text, forms, and tables from scanned documents	Amazon Textract
Pull totals and line items from invoices or receipts	Textract AnalyzeExpense
Read fields from driver licenses or passports	Textract AnalyzeID
Detect text in photos of scenes	Amazon Rekognition
Find sentiment, entities, or PII (Personally identifiable information)Information that can identify a person, alone or combined with other data, such as a name, email address, or passport number. Amazon Macie finds PII in Amazon S3.Glossary entry in the extracted text	Amazon Comprehend
Turn speech into text	Amazon Transcribe

Reverse mindmap: when the question points here

Is it Amazon Textract?

Yes, when the question says
- Extract text from scanned documents or PDFs leads to Textract
- Replace manual data entry from forms leads to Textract forms
- Keep the structure of tables leads to Textract tables
- Ask for a specific field in plain language leads to Textract Queries
- Invoices and receipts leads to AnalyzeExpense
- Driver licenses and passports leads to AnalyzeID
Know how it runs
- Single page, real time leads to Synchronous API
- Multipage PDF in S3 leads to Asynchronous API, completion to SNS
No, when the question says
- Text on a street sign in a photo leads to Rekognition
- Meaning or sentiment of text leads to Comprehend

Key takeaways

Textract extracts printed and handwritten text from documents, plus forms, tables, and specific fields.
AnalyzeDocument covers forms, tables, and Queries; AnalyzeExpense covers invoices and receipts; AnalyzeID covers U.S. identity documents.
Synchronous APIs handle single pages in near real time; asynchronous APIs handle multipage documents in S3, with completion published to SNS.
It is priced per page, depending on the features used; combine it with Comprehend to analyze the text.