What is on-device OCR?
On-device OCR converts visible text in an image or scanned page into machine-readable text using processing on the iPhone. It can support search, copying, accessibility, and field suggestions without uploading the page to a developer-operated recognition server. OCR does not preserve every part of the source: reading order, tables, columns, typography, handwriting, faint printing, and decorative layouts can be misread or rearranged. The original image or PDF remains the source of truth.
How to get more reliable OCR from an iPhone scan
- Flatten the page and use a contrasting, uncluttered background.
- Use diffuse light without glare, hard shadows, or screen reflections.
- Hold the camera parallel to the page and keep all four edges visible.
- Review crop, rotation, focus, and the smallest important text.
- Compare recognized totals, dates, parties, and identifiers with the source.
- Keep the original whenever the text may affect payment, safety, legal, medical, or compliance decisions.
What are structured fields after OCR?
Plain OCR produces text. Structured recognition tries to label parts of that text—such as an invoice total, due date, tracking number, model, serial number, nutrition value, or warning. The extra label can improve filtering and reminders, but it adds another interpretation step. A supported document can still have missing or incorrect fields when the layout is unfamiliar, the image is blurred, or OCR substituted a character.
When should you avoid relying on OCR alone?
Do not use OCR text alone when a wrong character, decimal point, unit, warning, date, or identifier could change a payment, deadline, safety action, medical decision, or legal obligation. Keep the source image, compare the relevant region at readable zoom, and ask the issuing organization when the document itself is unclear. OCR is a retrieval and drafting aid—not a certified transcription service.
What ScanMind can and cannot do
ScanMind uses Apple device-side frameworks for current English and Simplified Chinese OCR, classification, and supported field extraction. It preserves the source, lets you edit suggestions, and files confirmed records into a searchable library. It cannot promise perfect OCR, preserve every visual reading order, verify that a suggested field is correct, or turn a scan into a legal, medical, safety, or regulatory certification.
Apple documents text recognition through the Vision framework and camera-based document capture through VisionKit. ScanMind's exact supported categories and boundaries are listed on the structured recognition page.
How does ScanMind turn OCR into a reviewable record?
- The iPhone captures or imports the source page.
- Apple device-side recognition produces candidate text.
- Supported patterns suggest a document type and useful fields.
- The review screen keeps those suggestions editable.
- The user confirms, files, searches, or discards the record.
The source remains available for comparison. Suggested fields are conveniences for organization and search, not proof that the source says what the suggestion claims.
Primary technical sources
Related ScanMind guides
Use the printable scan checklist before capture, and read the private scanner guide before enabling sync or sharing sensitive documents.
Published August 2, 2026 · Updated and reviewed August 6, 2026 by Omuuz Editorial.