License Plate OCR Investigation System

A comprehensive license plate detection and OCR system for law enforcement investigations. Organized by projects, with multi-image cross-validation for improved accuracy.

Project Structure

Each investigation follows this structure:

projects/
├── 001/                    # Investigation 001
│   ├── raw/               # Source images of the vehicle
│   │   ├── image1.jpg     # Multiple angles/lighting
│   │   ├── image2.jpg     # of the same vehicle
│   │   └── image3.jpg
│   ├── annotations/       # Human annotations (JSON)
│   │   ├── image1.json    # License plate locations
│   │   └── image2.json
│   ├── debug/            # Debug images and parameters
│   │   ├── detection_parameters.json
│   │   └── [debug images]
│   ├── output/           # Analysis results
│   │   ├── investigation_report.json
│   │   ├── investigation_summary.txt
│   │   └── top_candidates.txt
│   └── project.json      # Project metadata
└── 002/                  # Investigation 002
    └── ...

Quick Start

1. Create a New Investigation

python project_manager.py create "Maryland license plate investigation - Case #2024-001"

This automatically creates the next project ID and folder structure.

2. Add Your Images

Copy your multiple images of the same vehicle to the projects/001/raw/ folder.

3. Annotate License Plates

python project_manager.py annotate 1

Draw bounding boxes around license plates in each image.

4. Run Comprehensive Analysis

python project_manager.py analyze 1

This runs the complete analysis pipeline:

• Automatic annotation analysis (generates detection parameters)
• Multi-image cross-validation
• Super resolution enhancement
• Character frequency analysis
• Combined ranking and reporting

Tools

Project Manager (project_manager.py)

• create "description": Create new project (auto-increments ID)
• annotate <id>: Run annotation tool (skips already annotated images)
• annotate <id> --reannotate: Reannotate all images (shows existing annotations)
• analyze <id>: Run comprehensive analysis (includes automatic annotation analysis)
• list: List all projects
• info <id>: Show project details

Annotation Tool (annotate_project.py)

• Interactive tool to mark license plate regions
• Generates detection parameters from annotations
• Use --analyze flag to process annotations

Detection System (detect_project.py)

• Comprehensive detection using multiple methods
• Saves debug images and results
• Generates ranked candidate list

Multi-Image Analysis (multi_image_analysis.py)

• Cross-validates candidates across multiple images
• Character pattern analysis for higher accuracy
• Specialized for multiple images of same vehicle

Super Resolution Analysis (super_resolution_analysis.py)

• Extracts annotated license plate regions
• Normalizes dimensions and applies super resolution
• Advanced enhancement techniques (EDSR, frequency domain, etc.)
• Enhanced OCR on super-resolved regions

Output Files

output/analysis_results.txt

Technical results summary with ranked candidates and frequency analysis.

output/comprehensive_analysis.json

Detailed JSON with all analysis results, including:

• Ranked candidates with scores
• Individual image results
• Cross-image analysis

output/super_resolution_report.json

Super resolution analysis results (if annotations available):

• Enhanced candidates from normalized regions
• Method breakdown and confidence scores
• Super resolution technique comparison

debug/detection_parameters.json

Optimized detection parameters (if annotations were provided).

Example Investigation Workflow

# Create investigation (auto-increments to next available project ID)
python project_manager.py create "Vehicle pursuit - Case #2024-001"

# Copy your images to the raw folder
# cp front_view.jpg rear_view.jpg side_view.jpg projects/001/raw/

# Annotate license plates
python project_manager.py annotate 1

# Run comprehensive analysis
python project_manager.py analyze 1

# Check final results
cat projects/001/output/analysis_results.txt

Requirements

• Python 3.7+
• OpenCV
• NumPy
• Pytesseract
• PIL (Pillow)

Notes

• The system uses aggressive detection by default
• Annotations improve detection accuracy significantly
• Debug images are saved for troubleshooting
• Results are automatically ranked by Maryland license plate likelihood