| --- |
| title: Chassis OCR |
| emoji: 🔧 |
| colorFrom: blue |
| colorTo: green |
| sdk: docker |
| pinned: false |
| --- |
| |
| # Chassis OCR Pipeline |
|
|
| Reads engraved chassis numbers from phone images and matches them against barcode-scanned numbers. |
|
|
| ## Setup |
|
|
| ```bash |
| pip install -r requirements.txt |
| ``` |
|
|
| On Linux you also need: |
| ```bash |
| sudo apt-get install libzbar0 |
| ``` |
|
|
| ## Project Structure |
|
|
| ``` |
| chassis_ocr/ |
| ├── images/ |
| │ ├── barcode/ ← put your 50 barcode images here |
| │ └── chassis/ ← put your 50 chassis images here (same filenames) |
| ├── results/ ← comparison images + report saved here |
| ├── preprocess.py ← image cleaning pipeline |
| ├── barcode_scanner.py |
| ├── ocr.py |
| └── evaluate.py ← run this |
| ``` |
|
|
| ## Important — Naming Convention |
|
|
| Barcode and chassis images must have the **same filename** to be paired: |
| ``` |
| images/barcode/001.jpg ←→ images/chassis/001.jpg |
| images/barcode/002.jpg ←→ images/chassis/002.jpg |
| ``` |
|
|
| ## Run |
|
|
| ```bash |
| # Test preprocessing on a single chassis image |
| python preprocess.py images/chassis/001.jpg |
| |
| # Test OCR on a single chassis image |
| python ocr.py images/chassis/001.jpg |
| |
| # Run full evaluation on all 50 pairs |
| python evaluate.py |
| ``` |
|
|
| ## Output |
|
|
| After running `evaluate.py`: |
| - `results/report.json` — full accuracy breakdown |
| - `results/*_comparison.jpg` — before/after preprocessing for each image |
|
|
| ## Pipeline |
|
|
| ``` |
| Barcode image → pyzbar → ground truth string |
| ↘ |
| Chassis image → CLAHE → Adaptive threshold Compare → ✅ Match / ❌ Mismatch |
| → Morphological ops → PaddleOCR ↗ |
| ``` |
|
|
