Arabic Text Image Segmentation

📋 Project Overview

This project focuses on extracting and merging word regions from an image using region-growing segmentation techniques. The process involves identifying regions, computing bounding boxes, measuring distances between regions, and merging close regions based on a threshold.

🚀 Features

• Region Extraction: Uses a region-growing algorithm to segment words in an image.
• Bounding Box Calculation: Computes bounding boxes for each detected region.
• Distance Measurement: Measures the distance between bounding boxes to determine closeness.
• Region Merging: Merges regions that are close to each other based on a specified threshold.
• Visualization: Draws bounding boxes around detected and merged regions for visualization.

📊 Libraries Used

• OpenCV: Advanced image processing
• NumPy: Matrix operations
• Matplotlib: Visualization

📦 Installation

# Clone the repository
git clone https://github.com/YacineAitKaci/Arabic-Text-Image-Segmentation.git

# Navigate to the project directory
cd Arabic-Text-Image-Segmentation

# Install required libraries
pip install numpy matplotlib opencv-python

🖼️ How to Use

1. Run the main segmentation notebook:

jupyter notebook main.ipynb

🛠️ Customization

• Adjust "seuil" and "seuil_seed" to modify the sensitivity of region extraction.
• Modify "fusion_seuil" to control how close two regions must be to merge.

⚡ Challenges Faced

• Selecting Seed Points: The choice of initial seed points greatly affects the segmentation results.
• Threshold Sensitivity: The method is highly sensitive to the threshold value, which can lead to under- or over-segmentation.
• Noise Sensitivity: Noisy images may cause false region growth, leading to inaccurate segmentation.
• Computation Time: Large images with many small regions can result in slow processing.
• Merging Criteria: Determining the correct criteria to merge regions without losing important details can be complex.

💡 Future Improvements

• Adaptive Thresholding: Implement an adaptive thresholding technique to dynamically adjust parameters based on the image characteristics.

• Machine Learning Integration: Use machine learning models to improve region classification and merging.

• Parallel Processing: Optimize the algorithm using parallel processing to improve performance on large images.

• Build OCR Model: Build and train a deep learning model for optical caracter recognition (OCR) using region-growing segmentation technique.

---