What is Amazon Textract?
Amazon Textract is a machine learning service that automatically extracts text, form data, and tables from scanned documents. Unlike simple OCR solutions, Textract understands document structure and can recognize relationships between form fields and their values.
The service solves the problem of manual document processing. Instead of manually typing invoices, forms, or contracts, Textract automatically extracts relevant information in structured form.
Core Features
- Text recognition (OCR) for printed and handwritten text
- Form extraction with automatic mapping of labels to values
- Table extraction preserving row and column structure
- Specialized APIs for invoices, IDs, and pay stubs
- Asynchronous processing for large document volumes
Typical Use Cases
Invoice Processing: Automatic extraction of invoice number, date, line items, and amounts from PDF invoices from various suppliers. Integration into accounting systems without manual data entry.
Contract Analysis: Extraction of key information from contracts such as parties, dates, amounts, and terms. Building searchable contract archives.
ID Verification: Automatic extraction of personal data from ID documents for KYC processes in banks and insurance companies.
Benefits
- No ML expertise required for usage
- Structured output with confidence scores
- Scales automatically with document volume
- Pay-per-use without base fees
Integration with innFactory
As an AWS Reseller, innFactory supports you with Amazon Textract: document processing workflow design, integration into existing systems, quality assurance of extraction results, and combination with other AWS services like Comprehend or Lambda.
Typical Use Cases
Frequently Asked Questions
What can Textract recognize?
Textract recognizes printed and handwritten text, form fields with key-value pairs, tables with rows and columns, and specific document types like invoices and IDs. Results are returned as structured JSON data.
Which document formats are supported?
Textract processes PDF documents (including multi-page) and image formats like JPEG, PNG, and TIFF. Synchronous processing is limited to one page, while asynchronous jobs can process several hundred pages.
How accurate is the text recognition?
Accuracy depends on document quality. For clear, printed documents, Textract achieves very high recognition rates. Handwriting and poor scans reduce accuracy. Textract returns confidence scores for each recognized element.
Can Textract process German documents?
Yes, Textract supports German and many other languages. Table and form recognition works language-independently as it is based on visual structures.