Iranian License Plate Character Segmentation

A Computer Vision project designed to isolate and segment the 8 characters of Iranian license plates. This project utilizes Morphological Operations and Geometric Heuristics to achieve high accuracy without the computational weight of Deep Learning models.

📌 Overview

Character Segmentation is a critical step in Automatic License Plate Recognition (ALPR) systems. This project addresses common challenges found in Iranian plates, such as:

• Shadows & Low Contrast: uneven lighting conditions.
• Blue Strip Interference: The standard blue flag/strip on the left.
• Connected Characters: Digits merging due to blur or low resolution.
• Physical Damage: Broken characters or screw/bolt noise.

Performance:

• Accuracy: 98.68% (Tested on 70+ challenging samples).
• Speed: Real-time processing using standard CPU.

🚀 Methodology

The algorithm follows a strict processing pipeline:

1. Preprocessing:

   • Resizing to a fixed height (80px).
   • Blue Strip Masking: Automatically whiting out the left 12% to prevent false positives from the flag.
   • TopHat Transform: Enhances contrast to handle heavy shadows.
   • Otsu Thresholding & Morphology: Converts to binary and cleans salt-and-pepper noise.

2. Segmentation & Heuristics:

   • Contour Extraction: Identifies potential character blobs.
   • Vertical Projection Profile: Detects valleys in pixel density to split characters that are "stuck" together.
   • Fragment Merging: Recombines broken character parts (e.g., the letter 'گ' or '5').
   • Geometric Filtering: Filters based on Aspect Ratio and Height relative to the plate.

3. Final Selection:

   • Median-Y Alignment: Calculates the central line of the plate and discards noise (like frames or screws) that falls too far from the center.
   • Sorting: Returns the top 8 candidates sorted from left to right.

📂 Project Structure

├── inputs/           # Directory containing raw license plate images
├── output/           # Resulting images with drawn bounding boxes
├── main.py           # Core algorithm implementation
├── requirements.txt  # Python dependencies
├── report.pdf        # Detailed project report (Persian)
└── README.md         # Project documentation

🛠️ Installation & Usage

1. Clone the repository:

   git clone https://github.com/mdvr9980/Iranian-License-Plate-Segmentation.git
   cd Iranian-License-Plate-Segmentation

Install dependencies:

pip install -r requirements.txt

Run the script:

Place your images in the inputs folder and run:

python main.py

Check Results:

Processed images with visual bounding boxes will be saved in the output directory. The terminal will display the number of characters found and the accuracy score for each image.

📊 Results

The algorithm successfully handles various edge cases:

Challenge	Solution
Shadows	TopHat Transformation highlights text regardless of lighting.
Blur/Sticking	Vertical Projection finds the optimal cut point between digits.
Blue Bar	Pre-masking the left margin eliminates false detections.
Noise	Median-Y alignment filters out screws and frame borders.