RecognizeDocumentsRequest

New Vision framework API announced at WWDC 2025. Extends text recognition to understand document structure — tables, lists, paragraphs, barcodes, and auto-detected data types. Available in iOS 19+, macOS 26+, iPadOS 19+, visionOS 3+.

Why It Matters for Project Aries

This is a potential game-changer for receipt scanning on iOS. It provides structured data extraction (tables → line items, Data Detection → dates/currency/phones) entirely on-device, removing the need for cloud OCR APIs in many use cases. It is the most significant iOS document scanning advancement since VNDocumentCameraViewController (iOS 13).

Capabilities

Document Structure

  • Tables: 2D cell array, row/column ranges, merged cell support, bounding regions
  • Lists: hierarchical items with nesting
  • Paragraphs: logical line grouping
  • Barcodes: QR codes and other machine-readable codes

Data Detection

Auto-identifies in text: email addresses, phone numbers, URLs, dates, measurements, currency, flight numbers, payment identifiers, tracking numbers, postal addresses.

Uses the DataDetection framework internally.

Text Access Methods

ViewDescription
transcriptAll text as single string
linesArray of lines
paragraphsLines grouped into paragraphs
wordsIndividual words (not CJK/Thai)
detectedDataMachine-identified data items

DocumentObservation Hierarchy

DocumentObservation
  └── document (Container)
       ├── text
       ├── tables → rows → cells → content (Container with text, etc.)
       ├── lists → items → content
       └── barcodes

Receipt Scanning Relevance

Table parsing directly maps to line items:

  • Each row = one purchased item
  • Cells = item name, quantity, unit price, line total
  • Data Detection identifies currency amounts in cells
  • Date detection identifies transaction date

26 languages covers major receipt languages.

Open Questions

  • Real-world accuracy on varied receipt types (thermal paper, crumpled, multi-column, handwritten) — untested at scale as of June 2026
  • Performance on older devices
  • Whether table detection reliably identifies receipt line-item layouts vs. other table types
  • The API was previewed at WWDC25 (2025); production availability expected with iOS 19/fall 2025 release cycle