Image Segmentation – DSA Project

Overview

This project implements an Image Segmentation System using concepts from Data Structures and Algorithms (DSA).
The segmentation is performed through a mean-similarity–based region merging algorithm supported by the Disjoint Set Union (Union–Find) data structure.

Users can upload an image through the frontend, and the backend processes it to generate a clean and meaningful segmented output.

---

Team Members — Group 31

• Manthan Nayak
• Sai Swaroop Guntuku
• Ankur Kumar

---

Objective

To segment an image into meaningful regions by grouping similar pixels and avoiding over-segmentation caused by noise, shadows, and natural variations in color.

---

Key Idea

Traditional pixel-wise comparison often results in noisy or fragmented segmentation.
To overcome this, the project focuses on region-level merging using:

• Mean color similarity
• Spatial neighborhood information
• Region-level comparison rather than pixel-level

The merging process is efficiently handled using the Union–Find (DSU) structure.

---

Tech Stack

Backend

• Python
• Flask
• NumPy
• OpenCV
• DSU (Union–Find)

Frontend

• HTML
• CSS
• JavaScript

---

Process

• Upload image through a simple UI
• Region-based segmentation using a DSU merging algorithm
• Displays:
  • Original Image
  • Intermediate Region Merging
  • Final Segmented Output
• Clean and responsive design

---

How to Run the Project

1. Install Dependencies

pip install numpy opencv-python flask

2. Start the Flask Backend

python app.py

3. Open the Frontend

Open the main HTML file from the frontend folder in your browser and connect it to the running backend.

4. Viewing Output

After processing:

• The Original Image will appear first.
• The Processed/Merged Image will appear next.
• Finally, the Segmented Output will be displayed with distinct colors.

Algorithm Summary

1. Extract pixel-level features:

   • Color information (R, G, B)
   • Spatial position (x, y)

2. Build edges between neighboring pixels:

   • 4-connectivity or 8-connectivity
   • Compute similarity between feature vectors

3. Initialize DSU (Disjoint Set Union):

   • Each pixel starts as its own set

4. Merge regions based on mean similarity:

   • Compare region means instead of individual pixels
   • Merge if similarity < threshold

5. Apply merging threshold to prevent over-merging:

   • Ensures objects do not collapse into one region

6. Generate final segmentation:

   • Assign unique labels to each region
   • Apply random/distinct colors for visualization

---

Folder Structure

DSA-PROJECT/
│── app.py
│── main.py
│── Frontend/
│   │── index.html
│   │── script.js
│   │── styles.css
│── uploads/
│── output/
│── README.md

---

Acknowledgements

This work was developed as part of the DSA Course Project (Group 31).

Conclusion

By integrating DSA concepts such as Union–Find with image processing techniques, this project produces clean, efficient segmentation results. It demonstrates the power of algorithmic thinking in solving real-world problems.

Thank you for exploring our project!