Problem → Solution → Impact
Problem: Few modern computer vision tools exist for identifying collectible coins.
Solution: This app uses machine learning and CV to recognise years and mint marks.
Impact: A working educational prototype for ML use cases in a niche hobby.
Rationale
A couple of friends in the US were manually identifying 1¢ coins under microscopes. This tool streamlines that: a desktop application that recognises years and mint marks on US cents using ML + CV.
Note: Data availability remains poor, so model accuracy is limited. It's a personal prototype, a fossil of fast coding and I'm okay with that.
Features at a Glance
| 🖼️ Image Handling | 🤖 Machine Learning |
|---|---|
| Image preprocessing tools | Basic classification model |
| Batch image processing | Option to train your own model |
| Format support: PNG, JPG... | OCR planned (future) |
| 💻 Desktop App UI | 📦 Export / Integration |
|---|---|
| Built with PySide6 (Qt) | Export results as CSV or JSON |
| Tabs for single/batch input | |
| Inline preview of results |
How It Works
Single Image Processing
- Click "Open Image" to load
- Tweak enhancement settings
- Click "Process"
- View the output in results panel
Batch Mode
- Switch to the Batch tab
- Choose multiple images or a folder
- Process them all at once
- Export results if needed
Image Requirements
| Condition | Details |
|---|---|
| Side of coin | Right side preferred |
| What must be visible | Year and mint mark |
| Image coverage | Partial coins are OK |
| Formats supported | PNG, JPG, JPEG, BMP, TIF, TIFF |
Training a Custom Model
You can improve results by training on your own dataset.
- Collect labelled coin images
- Place them in
data/raw - Use the Train Model tool in-app
- Follow the wizard to generate your new model
Future Improvements
- Transfer learning via deeper networks
- Text-based OCR for better year/mint detection
- Support for coins outside the US
- Lightweight mobile version