Photo Identification

Shelf-School treats a photograph as input to an identifier decoder, not as the library record itself.

Storage behavior

ISBN-intake and circulation images are temporary request data and are discarded after decoding. Available Open Library covers are downloaded when the book is created. A no-ISBN intake image is stored only when staff supplies a manual title and intends that image to be the cover.

Current pipeline

1. The browser optionally tries BarcodeDetector for immediate feedback.
2. The uploaded file is sent to Django.
3. Pillow opens the image locally.
4. zxing-cpp decodes supported barcodes and QR codes.
5. A COPY:<uuid> payload resolves directly to one copy.
6. An ISBN resolves through cataloged books and eligible copies.
7. The user reviews the result before confirming circulation.

For checkout, an ISBN candidate must have exactly one available matching copy. For return, it must have exactly one checked-out matching copy with an active loan. Ambiguous ISBNs require the unique copy QR.

Future multi-book recognition

An optional recognizer may eventually accept one scene containing several books and propose catalog matches. A safe implementation must:

• Keep the existing local barcode path as the default.
• Make paid or remote providers optional and explicit.
• Show confidence and provenance for every proposed match.
• Reject unknown or ambiguous physical copies.
• Require review before creating any loan.
• Preserve one Loan record per Copy.

Cover recognition alone cannot reliably identify which physical copy is being borrowed when several copies share the same title.