https://github.com/jjshay/handwriting-analysis
Forensic signature and handwriting analysis with statistical comparison
https://github.com/jjshay/handwriting-analysis
ai analysis authentication computer-vision forensics handwriting machine-learning opencv python signature-verification statistics
Last synced: 5 months ago
JSON representation
Forensic signature and handwriting analysis with statistical comparison
- Host: GitHub
- URL: https://github.com/jjshay/handwriting-analysis
- Owner: jjshay
- License: mit
- Created: 2026-01-08T20:01:14.000Z (6 months ago)
- Default Branch: main
- Last Pushed: 2026-01-12T05:25:41.000Z (5 months ago)
- Last Synced: 2026-01-12T06:00:00.281Z (5 months ago)
- Topics: ai, analysis, authentication, computer-vision, forensics, handwriting, machine-learning, opencv, python, signature-verification, statistics
- Language: Python
- Size: 108 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Security: SECURITY.md
Awesome Lists containing this project
README
# Handwriting Analysis System
[](https://codecov.io/gh/jjshay/handwriting-analysis)
**Forensic-grade signature and handwriting analysis - statistical comparison for authentication.**
[](https://asciinema.org/a/bWZcm0S3ayeCsVy4)
[](https://asciinema.org/a/bWZcm0S3ayeCsVy4)
---
## What Does This Do?
Analyze signatures and handwriting for authenticity:
1. **Extract features** from signature images (slant, pressure, proportions)
2. **Build baseline** from known authentic samples
3. **Compare questioned documents** against the baseline
4. **Generate reports** with statistical confidence scores
**Use cases:**
- Art authentication (artist signatures)
- Document verification
- Forensic analysis
- Research and education
---
## Quick Start
```bash
# Clone the repo
git clone https://github.com/jjshay/handwriting-analysis.git
cd handwriting-analysis
# Install dependencies
pip install -r requirements.txt
# Run the interactive demo
python demo.py
# Or run the visual showcase
python showcase.py
# Analyze with sample signatures
python forensic_handwriting_analyzer.py \
--baseline examples/sample_baseline_data.json \
--questioned examples/signature_questioned.png
```
### Sample Files
- `examples/signature_authentic_01.png` - Sample authentic signatures
- `examples/signature_questioned.png` - Questioned signature to analyze
- `examples/sample_baseline_data.json` - Pre-computed baseline stats
- `sample_output/forensic_report.txt` - Example analysis report
---
## Architecture
```mermaid
flowchart TB
subgraph Baseline["Baseline Creation"]
A[Authentic Samples] --> B[Image Preprocessing]
B --> C[Feature Extractor]
C --> D[Statistical Analyzer]
D --> E[Baseline Model]
end
subgraph Analysis["Document Analysis"]
F[Questioned Document] --> G[Image Preprocessing]
G --> H[Feature Extractor]
end
subgraph Comparison["Statistical Comparison"]
E --> I[Feature Comparison]
H --> I
I --> J[Z-Score Calculator]
J --> K[Mahalanobis Distance]
K --> L[Probability Engine]
end
subgraph Optional["Optional: AI Review"]
H --> M1[Claude]
H --> M2[GPT-4]
H --> M3[Gemini]
H --> M4[Grok]
M1 & M2 & M3 & M4 --> N[AI Consensus]
end
subgraph Output
L --> O[Confidence Score]
N --> O
O --> P[Forensic Report]
end
style A fill:#e1f5fe
style F fill:#fff9c4
style P fill:#c8e6c9
```
## How It Works
```
Known Authentic Samples (70+)
│
▼
┌─────────────────────────────────────────┐
│ FEATURE EXTRACTION │
│ - Slant angle (60-90°) │
│ - Height ratios │
│ - Stroke characteristics │
│ - Pressure patterns │
│ - Connected vs lifted strokes │
└─────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────┐
│ STATISTICAL BASELINE │
│ Mean ± Standard Deviation for each │
│ feature across all authentic samples │
└─────────────────────────────────────────┘
│
▼
Questioned Document
│
▼
┌─────────────────────────────────────────┐
│ STATISTICAL COMPARISON │
│ - Z-scores for each feature │
│ - Mahalanobis distance │
│ - Probability calculations │
└─────────────────────────────────────────┘
│
▼
Authentication Report
(Confidence: 94.7%)
```
---
## Features Extracted
| Feature | Description | Typical Range |
|---------|-------------|---------------|
| Slant Angle | Degrees from horizontal | 60-90° |
| Height Ratio | Full height / x-height | 2.0-3.5 |
| Aspect Ratio | Width / Height of bounding box | 1.5-4.0 |
| Stroke Width | Average pen stroke thickness | 2-8 px |
| Pressure Variance | Variation in stroke darkness | 0.1-0.4 |
| Connectedness | % of connected strokes | 40-80% |
| Loop Ratio | Loops per letter | 0.2-0.6 |
---
## Output Report
```
FORENSIC HANDWRITING ANALYSIS REPORT
=====================================
Document ID: QD-2024-001
Analysis Date: January 8, 2024
FEATURE COMPARISON:
Slant Angle: 73.2° (Baseline: 71.5° ± 4.2°) → WITHIN RANGE
Height Ratio: 2.8 (Baseline: 2.6 ± 0.3) → WITHIN RANGE
Stroke Width: 4.1px (Baseline: 3.8 ± 0.8) → WITHIN RANGE
Pressure Var: 0.28 (Baseline: 0.25 ± 0.08) → WITHIN RANGE
STATISTICAL ANALYSIS:
Combined Z-Score: 0.82
Mahalanobis Distance: 1.24
CONCLUSION:
Confidence: 94.7%
Opinion: The questioned signature is CONSISTENT with
the known authentic baseline samples.
```
---
## Multi-AI Review (Optional)
For additional verification, the system can use multiple AI models:
| AI Model | Role |
|----------|------|
| Claude | Detailed stroke analysis |
| GPT-4 | Pattern recognition |
| Gemini | Visual comparison |
| Grok | Anomaly detection |
Each AI provides independent analysis, then results are combined for consensus.
---
## Setup
### Basic (Statistical Only)
```bash
pip install -r requirements.txt
python demo.py
```
### Full AI Mode
```bash
cp .env.example .env
nano .env
# Add API keys
python llm_handwriting_reviewer.py
```
---
## Files
| File | Purpose |
|------|---------|
| `forensic_handwriting_analyzer.py` | Main statistical analyzer |
| `signature_analyzer.py` | Signature-specific analysis |
| `statistical_signature_generator.py` | Generate test signatures |
| `llm_handwriting_reviewer.py` | Multi-AI review system |
| `demo.py` | Demo without dependencies |
---
## Important Notes
- This is a research/educational tool
- Not a replacement for professional forensic examination
- Statistical analysis provides probability, not certainty
- Always consult qualified experts for legal matters
---
## License
MIT - For research and educational use