CoinCheck Vision

Privacy-first visual screening for coins, bullion, and numismatic items.

CoinCheck Vision is a Python-based, on-premises visual screening platform for professional numismatic and bullion dealers, grading room staff, and lot-processing teams. It delivers a fast, repeatable first-pass identification layer — without sending images to a cloud service, without per-scan subscription fees, and without surrendering provenance data to a third party.

The system runs entirely on your own hardware (macOS or Linux, Apple Silicon or x86-64). A single laptop or workstation can screen hundreds of items per day, generate formatted due-diligence reports, and produce publication-ready auction listing text — all from the same command-line interface.

Screening limitation: CoinCheck Vision is a visual screening and due-diligence tool. It does not authenticate metal content, weight, or genuineness of any coin or bullion item. All outputs use "visual match confidence" language and are intended to supplement, not replace, physical testing and expert grading.

At a Glance

The Problem It Solves

A professional dealer processing a raw estate lot or incoming consignment faces a recurring bottleneck: every item needs a quick plausibility check before committing grading fees, shipping, or catalog time.

Current options are slow or require network round-trips:

• Manual lookup — experienced eye, but not scalable for high-volume lots
• Cloud vision APIs — coin-generic, privacy concerns, per-call billing, require internet
• PCGS/NGC photo lookup — cert-number dependent; useless for raw or uncertified material

CoinCheck Vision fills the gap: rapid, offline, calibrated visual identification against a reference database covering ancient denarii, modern bullion, error coins, banknotes, and fractional rounds — all in a single tool.

Core Capabilities

Single-Coin Identification

Match a photograph against the reference database using a multi-signal pipeline:

• ORB feature keypoint matching (weighted 75%)
• Numista metadata score (weighted 25%)
• Diameter, purity, and expected-text cross-checks
• Outputs confidence score (0.00–1.00), visual match tier, and full metadata

Accepts photos from any phone or camera — iPhone HEIC, Android AVIF, Google Pixel DNG, eBay WebP, flatbed TIFF, or standard JPEG. No format conversion required.

Batch Processing

Process an entire folder of coin images in one command:

• Per-coin JSON or CSV with match results
• Aggregate PDF or DOCX summary report
• Lot-level XML manifest for ERP or catalog ingestion

A 200-image lot typically completes in under 3 minutes on an M-series MacBook.

Listing Assistant

Automatically generates auction-ready listing text from an identification result:

• Long-form description with provenance language
• Purity, weight, diameter, and denomination populated from reference data
• Output in plain text, JSON, or formatted DOCX — ready to paste into HiBid, eBay, or internal auction software

Due-Diligence PDF Report

Single-coin or batch report in PDF or DOCX:

• Identification result with confidence score and disclaimer language
• Spot-price melt value estimate (AU/AG/PT/PD, updated every 15 minutes from a licensed data feed)
• Suitable for lot documentation, consignment intake, or client-facing screening records

OCR Slab and Legend Reading

Reads visible text from holder labels and coin legends (Latin, English, Arabic) using Tesseract OCR. Cross-references extracted text against the reference database for additional match signal — useful for slabbed material where label data supplements the image.

Barcode + NGC Cert Lookup

Decodes QR/barcode from slab images and performs a live NGC Population Report lookup, returning cert grade, population data, and holder type. Works for NGC-holdered ancient and modern coins.

How the Confidence Score Works

A HIGH confidence score means the photograph closely resembles the reference. It does not prove the coin is genuine, authenticate metal content, or replace physical testing. This limitation is printed on every report.

The system automatically raises named risk flags when conditions warrant closer review — including low match confidence, insufficient reference images, OCR legend mismatch, or single-image reference products.

Reference Database Coverage

987 calibrated product entries across 16 categories:

Roman series highlight: 39 Roman coins covering the full debasement arc — Augustus (~.97 fine silver denarius) through Crisis-era billon antoniniani (~.02–.05 fine) and Diocletian's 294 AD reform argenteus (~.95 fine). Purity metadata calibrated to numismatic scholarship.

Expandable: New product entries are JSON files with 22 standardized metadata fields. Dealers can contribute proprietary reference images for house-specialty coins, improving match accuracy for their specific inventory mix without affecting the shared database.

Dealer Use Case: Consignment Lot Triage

A consignment of 180 raw coins arrives for lot processing:

1. Photograph each coin (obverse + reverse)
2. Run batch identify on the folder — approximately 3 minutes for 180 coins
3. Review the PDF triage report — high-confidence matches flow directly to catalog prep; low-confidence items are flagged for desk review
4. Generate listing drafts for the high-confidence subset
5. Attach the due-diligence PDF to the consignment record for provenance documentation

Conservative time saving estimate:

At a $50/hr labor rate, a 180-coin lot saves approximately $650 per consignment in intake and cataloging labor.

Additional use cases:

• Grading room pre-screen — flag items warranting closer review before submitting to a grading service
• Estate auction triage — rapid classification of mixed lots to estimate realized-value ranges
• Dealer-to-dealer buy lists — consistent, documented screening results for negotiation
• Staff training — reproducible reference database for in-house training without requiring senior staff at every step

System Requirements

• macOS 13+ (Apple Silicon M1/M2/M3/M4 recommended) or Ubuntu 22.04+
• Python 3.11+
• 4 GB RAM minimum; 8 GB recommended for batch jobs
• All 8 image formats supported out of the box — no ImageMagick or additional installs required
• Tesseract OCR optional (for slab and legend reading)
• Docker 24+ optional (for REST API integration)
• No internet required for identify, batch, listing, or report commands

REST API Integration

An optional FastAPI wrapper ships as a Docker image for integration with existing dealer management systems, HiBid platforms, or auction back-office software:

Pilot Program

CoinCheck Vision is currently in pre-commercial beta. Pilot partnerships are available for qualified professional dealers under a fixed-fee evaluation agreement.

What we ask of pilot partners:

• Provide 50–100 photographed reference coins from house inventory to calibrate accuracy
• Share aggregate accuracy metrics (no individual coin data required)
• Participate in 2–3 structured feedback sessions over the pilot period

Interested in a pilot evaluation? Contact us

View Public Reference Catalogue

Disclaimer

CoinCheck Vision is a visual screening and due-diligence tool. It is not authentication, grading, appraisal, assay, valuation, financial advice, tax advice, legal advice, or investment advice. Visual image matching cannot verify metal content, purity, weight, diameter, thickness, density, edge reeding, authenticity, grade, or value. A HIGH confidence score means the photograph closely resembles the reference image — it does not prove the coin is genuine or certify any physical property. Physical verification by qualified numismatists, dealers, graders, or assayers is required before any purchase, sale, or consignment decision. Copyright © 2026 Viking FinTech Trading LLC. All rights reserved.