Amazon Textract: Intelligent Document Processing (IDP), Forms, Tables & Specialized APIs
Key Takeaways
Amazon Textract is a fully managed, serverless Intelligent Document Processing (IDP) service that goes beyond traditional Optical Character Recognition (OCR). It extracts printed text, handwriting, key-value pairs (forms), structured tables, signatures, and identity attributes from scanned documents and PDFs without requiring custom machine learning models.
Unlike basic OCR engines that produce flattened strings of text, Amazon Textract preserves the structural context and spatial geometry of documents. It provides specialized purpose-built APIs like AnalyzeID (driver's licenses and passports), AnalyzeExpense (invoices and receipts), and AnalyzeLending (mortgage packets).
Main Discussion
Amazon Textract API & Feature Breakdown
| API / Feature | Primary Capability | Key Output Structures | Target Use Case |
|---|---|---|---|
DetectDocumentText | Basic OCR for printed and handwritten text. | PAGE, LINE, WORD | Transcribing unstructured articles, letters, or plain text books. |
AnalyzeDocument (FORMS) | Extracts form fields as semantic key-value relationships. | KEY_VALUE_SET blocks linking label keys to their values | Processing tax forms, loan applications, and medical intake records. |
AnalyzeDocument (TABLES) | Extracts tabular data while preserving row, column, and cell relationships. | TABLE, CELL blocks with RowIndex and ColumnIndex | Extracting balance sheets, stock market reports, and multi-row inventory tables. |
AnalyzeDocument (QUERIES) | Uses natural language questions to retrieve specific data points regardless of layout changes. | Specific answer text and confidence score | Asking "What is the policy expiration date?" without pre-defining bounding coordinates. |
AnalyzeDocument (SIGNATURES) | Detects the presence and location of handwritten or electronic signatures. | Signature bounding box and confidence score | Verifying whether contracts, waivers, or consent forms have been signed. |
AnalyzeID | Specialized pre-trained parser for identity credentials. | Normalized fields: FIRST_NAME, LAST_NAME, ID_TYPE, EXPIRATION_DATE | KYC identity verification during bank onboarding or car rental checkouts. |
AnalyzeExpense | Pre-trained extraction for financial receipts and invoices. | VendorName, InvoiceTotal, DueDate, line-item tables | Automated accounts payable workflows and corporate expense reporting. |
The Textract Block Graph Hierarchy
Amazon Textract does not output raw strings; it returns an array of structured Block objects connected as a directed graph:
- Geometry & Bounding Box: Every block contains coordinate geometry (
BoundingBox: Left, Top, Width, HeightandPolygonpoints) relative to the document page. - Confidence Score: Each word, key-value pair, and table cell includes a machine learning confidence score () to assist with automated routing to human review via Amazon Augmented AI (Amazon A2I).
Synchronous vs. Asynchronous Processing Modes
| Mode | API Calling Pattern | Supported Formats & File Limits | Ideal Workload |
|---|---|---|---|
| Synchronous | DetectDocumentTextAnalyzeDocumentAnalyzeExpenseAnalyzeID | • Single-page PNG, JPEG, TIFF • Direct base64 bytes or S3 object (up to 5 MB) | Real-time user uploads, mobile document scanning, interactive web apps. |
| Asynchronous | StartDocumentAnalysis / GetDocumentAnalysisStartDocumentTextDetection / GetDocumentTextDetection | • Multi-page PDF, TIFF • Documents stored in Amazon S3 (up to 500 MB / 3,000 pages) | Scheduled batch processing of mortgage portfolios, legal document archives. |
Service Disambiguation: Textract vs. Rekognition vs. Comprehend
- Amazon Textract: Parses scanned documents, PDFs, forms, and tables to preserve document layout and relationships.
- Amazon Rekognition: Detects text embedded in natural scene photographs (e.g., street signs, race bibs, billboard photos).
- Amazon Comprehend: Performs NLP comprehension (sentiment, language, standard entities, PII) on digital text strings extracted by Textract.
Exam Guide
Exam Tips
- Primary Service Purpose: Amazon Textract is the default service for Intelligent Document Processing (IDP), extracting text, forms, tables, and handwritten data from scanned documents and PDFs.
- Feature Triggers:
- Extracting key-value form fields (e.g., "Name: John")
AnalyzeDocumentwithFORMS. - Extracting multi-column tables and cell data
AnalyzeDocumentwithTABLES. - Parsing driver's licenses or passports
AnalyzeIDAPI. - Extracting line items, vendor names, and totals from receipts/invoices
AnalyzeExpenseAPI. - Asking specific questions about document contents regardless of layout shifts
AnalyzeDocumentwithQUERIES.
- Extracting key-value form fields (e.g., "Name: John")
- Multi-Page PDFs: Processing multi-page PDFs requires Asynchronous APIs (
StartDocumentAnalysis/GetDocumentAnalysis) with files stored in Amazon S3. - Human-in-the-Loop Integration: Amazon Textract integrates with Amazon Augmented AI (Amazon A2I) to send low-confidence document extractions to human reviewers.
Practice Test
Question 1
A loan origination company needs an automated solution to extract customer names, employment history, and financial table data from thousands of multi-page scanned PDF application forms. The solution must preserve the relationship between form labels and their associated values, as well as extract table cells into rows and columns with minimal operational overhead. Which AWS service should be used?
- A. Amazon Rekognition Text in Image
- B. Amazon Textract with Forms and Tables analysis features enabled
- C. Amazon Comprehend Syntax Tokenizer
- D. Amazon Polly Speech Marks
Correct Answer
- B. Amazon Textract with Forms and Tables analysis features enabled
- Explanation: Amazon Textract with the Forms and Tables features enabled (
AnalyzeDocument) is specifically designed to extract structured key-value pairs from forms and preserve row-and-column structures from tabular data in scanned PDF documents.
- Explanation: Amazon Textract with the Forms and Tables features enabled (
Question 2
A fintech mobile application requires newly registered users to upload a photo of their government-issued driver's license or passport. The application must automatically extract the user's date of birth, document expiration date, and ID number as standardized, normalized fields. Which Amazon Textract API is designed for this use case?
- A.
DetectDocumentText - B.
AnalyzeID - C.
AnalyzeExpense - D.
DetectModerationLabels
Correct Answer
- B.
AnalyzeID- Explanation: The
AnalyzeIDAPI in Amazon Textract is specialized for extracting normalized fields (such as document number, expiration date, and date of birth) from government-issued identity documents, including driver's licenses and passports.
- Explanation: The
Question 3
A legal firm is digitizing a large volume of contracts and other legal documents to streamline document management and analysis. The firm is looking for AWS machine learning services that can automatically detect and extract text from scanned images and photos of these documents. Understanding which services specialize in text extraction from images will help the firm choose the right solution for automating its document processing workflow.
Which AWS ML services will you recommend to the firm? (Select two)
- A. Amazon Rekognition
- B. Amazon Polly
- C. Amazon Textract
- D. Amazon Lex
- E. Amazon Comprehend
Both Rekognition and Textract possess the ability to detect text within images:
Correct Answers
- A. Amazon Rekognition
- Explanation: Rekognition specializes in identifying text located spatially within an image, for instance, words displayed on street signs, t-shirts, or license plates. Its typical use cases encompass visual search, content filtering, and deriving insights from content. However, it is not the ideal choice for images containing more than 100 words, as this exceeds its limitation.
- C. Amazon Textract
- Explanation: Textract is tailored specifically towards processing documents and PDFs, offering a comprehensive suite for Optical Character Recognition (OCR). It accurately extracts text, forms, and tables from financial reports, medical records, receipts, ID documents, and legal contracts.
Incorrect Answers
- B. Amazon Polly
- Explanation: Amazon Polly uses deep learning technologies to synthesize natural-sounding human speech from text, converting written documents into audio.
- D. Amazon Lex
- Explanation: Amazon Lex is a fully managed artificial intelligence service with advanced natural language models designed to build conversational interfaces (chatbots) and cannot extract text from documents or images.
- E. Amazon Comprehend
- Explanation: Amazon Comprehend is a natural language processing (NLP) service that extracts insights, sentiment, key phrases, and entities from already-extracted text, but it cannot extract text from documents or images directly.
Reference: