Satellite Image Feature Extraction System

Road & Building Classification using Digital Image Processing

Cairo University - Faculty of Engineering
Computer Engineering Department
Digital Image Processing Lab - Term Project 2025

---

📋 Project Overview

This project implements a complete satellite image processing pipeline for detecting and classifying roads and buildings from aerial/satellite imagery. The system uses traditional digital image processing techniques (no deep learning in the main pipeline) and outputs results in both raster (PNG) and vector (SVG) formats.

Key Features

• ✅ SLIC superpixel segmentation for boundary-preserving regions
• ✅ Region merging to reduce over-segmentation
• ✅ Road-focused feature extraction (skeleton, linearity, color uniformity)
• ✅ Adaptive soft-scoring classification (no hard thresholds)
• ✅ Morphological cleanup for connected road networks
• ✅ SVG vectorization using Douglas-Peucker line simplification (from scratch)
• ✅ User-friendly GUI application

---

🗂️ Project Structure

Imaging/
├── Module1/                    # Input images (satellite/aerial imagery)
├── Module2/
│   ├── color/                  # Preprocessed color images
│   └── gray/                   # Preprocessed grayscale images
├── Module3/
│   ├── edges/                  # Edge detection results
│   └── segmentation/           # SLIC segmentation results
├── Module4/
│   ├── region_features/        # Extracted features per region
│   └── region_clusters/        # Classification results
├── Module5/
│   └── visualization/          # Final PNG visualizations with overlays
├── Module6/                    # Final SVG vector outputs ⭐
├── Module_DL/                  # Deep Learning outputs (optional)
│
├── phase1_preprocessing.py     # Phase 1: Preprocessing (resize, denoise, enhance)
├── phase2_improved.py          # Phase 2: SLIC Segmentation
├── phase3_improved.py          # Phase 3: Feature Extraction
├── phase4_improved.py          # Phase 4: Classification
├── phase5_improved.py          # Phase 5: Visualization
├── phase6_vectorization.py     # Phase 6: SVG Vectorization ⭐
├── phase_dl_classification.py  # Deep Learning comparison (optional) 🧠
├── gui_app.py                  # GUI Application ⭐
├── run_improved_pipeline.py    # Pipeline runner script
└── README.md                   # This file

---

🚀 Installation & Requirements

Prerequisites

• Python 3.8 or higher (tested on Python 3.12)
• Operating System: Windows, Linux, or macOS

Required Libraries

Install all dependencies using pip:

pip install numpy opencv-python scikit-image scikit-learn scipy pillow

Complete Library List:

Library	Version	Purpose
numpy	>= 1.21.0	Array operations and numerical computing
opencv-python	>= 4.5.0	Image processing (cv2)
scikit-image	>= 0.18.0	SLIC superpixels, morphological operations
scikit-learn	>= 0.24.0	K-means clustering, StandardScaler
scipy	>= 1.7.0	Scientific computing, ndimage operations
pillow	>= 8.0.0	Image display in GUI (PIL/ImageTk)

GUI Requirements (Tkinter):

• Windows: Tkinter is included with Python by default
• Linux (Ubuntu/Debian):

  sudo apt-get install python3-tk

• macOS: Tkinter is included with Python by default

Virtual Environment Setup (Recommended)

# Create virtual environment
python -m venv venv

# Activate virtual environment
# Windows:
venv\Scripts\activate
# Linux/macOS:
source venv/bin/activate

# Install dependencies
pip install numpy opencv-python scikit-image scikit-learn scipy pillow

---

📖 How to Use

Option 1: GUI Application (Recommended) ⭐

Step 1: Activate your virtual environment (if using one)

# Linux/macOS
source venv/bin/activate

# Windows
venv\Scripts\activate

Step 2: Run the GUI application

python gui_app.py

Step 3: Using the GUI

1. Select an image from the list on the left panel (images from Module1 folder)
2. Click "▶ Start Processing" button in the right panel
3. Wait for all 6 phases to complete (green checkmarks indicate success)
4. View results using the tabs: Overlay, Roads, Buildings
5. Click "📐 Open SVG Output" to open the vector file
6. Click "📂 Open Folder" to browse all output files

Option 2: Command Line Pipeline (All Images)

Process all images in Module1 at once:

python run_improved_pipeline.py

This will:

1. Process all .jpg, .png, .jpeg, .tif images in Module1
2. Run all 6 phases sequentially
3. Save outputs to Module5 (visualizations) and Module6 (SVG files)

Option 3: Process Single Image (Command Line)

Process a specific image by running each phase with the --image flag:

python phase2_improved.py --image your_image.jpg
python phase3_improved.py --image your_image.jpg
python phase4_improved.py --image your_image.jpg
python phase5_improved.py --image your_image.jpg
python phase6_vectorization.py --image your_image.jpg

Option 4: Individual Phases (All Images)

Run each phase separately on all images:

python phase2_improved.py   # SLIC Segmentation
python phase3_improved.py   # Feature Extraction
python phase4_improved.py   # Classification
python phase5_improved.py   # Visualization
python phase6_vectorization.py   # SVG Vectorization

---

⚡ Quick Start

1. Place your satellite/aerial images in the Module1/ folder
2. Run python gui_app.py
3. Select an image and click "Start Processing"
4. Find your results in Module6/ (SVG) and Module5/visualization/ (PNG)

---

🔧 Troubleshooting

Common Issues

Issue	Solution
ModuleNotFoundError: No module named 'cv2'	Run pip install opencv-python
ModuleNotFoundError: No module named 'skimage'	Run pip install scikit-image
_tkinter.TclError: no display name	Set display: export DISPLAY=:0 (Linux)
TclError: bad argument "zoomed"	Already fixed - update to latest code
Images not appearing in list	Check that images are in Module1/ folder
SVG file not opening	Install SVG viewer or use a web browser

Linux-Specific Setup

If running on Linux, ensure you have:

# Install Tkinter
sudo apt-get install python3-tk

# Install xdg-utils for opening files
sudo apt-get install xdg-utils

---

📊 Pipeline Phases

Phase 1: Preprocessing (phase1.ipynb)

• Converts input images to color and grayscale versions
• Applies brightness/contrast normalization (GUI handles this)

Phase 2: Segmentation (phase2_improved.py)

• Uses SLIC superpixels for boundary-respecting segmentation
• Merges similar regions based on LAB color distance
• Removes small noise regions

Phase 3: Feature Extraction (phase3_improved.py)

• Skeleton features: Length, ratio, linearity (key for roads)
• Shape features: Compactness, elongation, rectangularity, solidity
• Color features: Saturation, gray-ness, uniformity
• Edge features: Internal edge density, boundary strength

Phase 4: Classification (phase4_improved.py)

• Soft-scoring approach for roads and buildings
• Roads: High elongation, high linearity, low saturation
• Buildings: High compactness, high rectangularity, small area

Phase 5: Visualization (phase5_improved.py)

• Morphological cleanup of road networks
• Connected component analysis
• Overlay generation with legend

Phase 6: Vectorization (phase6_vectorization.py)

• Extracts contours from classification masks
• Applies Douglas-Peucker line simplification (implemented from scratch)
• Generates SVG files compatible with mapping tools

---

🎯 Output Formats

PNG Visualizations (Module5/visualization/)

• {image}_overlay.png - Classification overlay on original
• {image}_overlay_with_legend.png - With classification legend
• {image}_roads_only.png - Binary road mask
• {image}_buildings_only.png - Binary building mask

SVG Vectors (Module6/)

• {image}.svg - Scalable vector graphics with roads & buildings
• {image}_vectorized_overlay.png - Vector overlay preview

---

🛠️ Key Algorithms (Implemented from Scratch)

Algorithm	File	Description
Douglas-Peucker	phase6_vectorization.py	Line simplification for vector output
Region Merging	phase2_improved.py	Merge similar adjacent superpixels
Skeleton Features	phase3_improved.py	Road linearity detection
Soft Scoring	phase4_improved.py	Adaptive classification without hard thresholds
Morphological Filter	phase5_improved.py	Road network cleanup

---

📈 Classification Criteria

Roads

• ✅ High elongation (> 2.5)
• ✅ High linearity (long skeleton relative to area)
• ✅ Low saturation (gray colors)
• ✅ High color uniformity
• ✅ Moderate consistent width

Buildings

• ✅ High rectangularity (> 0.5)
• ✅ High compactness (circular shape factor > 0.25)
• ✅ High solidity (convex shape)
• ✅ Low elongation (< 3.0)
• ✅ Smaller area (< 6000 pixels)

---

👥 Team Members

Name	Role
[Team Member 1: mohamed emad]	Segmentation & Preprocessing
[Team Member 2: mohamed yasser nabil]	Feature Extraction
[Team Member 3: mazen hatem]	Classification & ML
[Team Member 4: karim yasser]	Visualization & GUI

---

📚 References

1. Garg, R.D. (2024). Integrated approaches for road extraction and de-noising from high resolution satellite image using adaptive global thresholding and morphological operations.

2. Chen, Z. et al. (2022). Road extraction in remote sensing data: A survey.

3. Pang, Z. et al. (2024). A Building Extraction Method for High-Resolution Remote Sensing Images.

4. Achanta, R. et al. (2012). SLIC Superpixels compared to state-of-the-art superpixel methods. IEEE TPAMI.

5. Douglas, D.H. & Peucker, T.K. (1973). Algorithms for the reduction of the number of points required to represent a digitized line.

---

⚠️ Notes

• This project uses no deep learning in the main pipeline (as per project requirements)
• A pretrained DNN comparison can be added for bonus marks
• All non-primitive functions are attributed in the code
• Output SVG is compatible with GIS/mapping tools

---

📄 License

This project is for educational purposes only (Cairo University DIP Lab 2025).