Amazon Textract
Amazon Textract is a machine learning service from Amazon Web Services that extracts text, handwriting, form fields and table data out of scanned documents. It goes beyond optical character recognition by preserving the structure it finds, so a key and its value, or a table cell and its column, come back related rather than as loose words.
AI Practitioner focus
- Textract extracts printed/handwritten text plus document structure such as key-value pairs, tables, queries, invoices/receipts, and identity/lending fields.
- It is OCR/document intelligence, not NLP sentiment/entities (Amazon Comprehend) or general image recognition (Amazon Rekognition). Human review can handle uncertain results.
Key points
- Reads printed text, handwriting, forms and tables from images and PDFs, returning the extracted values along with their relationships.
- Specialised extraction for common document types: invoices, receipts, identity documents and lending or mortgage packets.
- Removes the need for manual data entry and template-based scraping when digitising paperwork.
- Requires no machine learning expertise — it is a pre-trained service accessed through an API, unlike Amazon Sage Maker, where you build and train the model yourself.
- The boundary that matters: Textract pulls text out of a document image. Amazon Comprehend then interprets that text — sentiment, key phrases, entities, language. The pair is often used in sequence. Amazon Rekognition analyses images for objects, scenes and faces rather than for document text.