Image Segmentation Using K-Means & PCA – README

This project demonstrates how to perform image segmentation and dimensionality reduction using:

• K-Means Clustering (for color-based segmentation)
• PCA (Principal Component Analysis) (for image compression)

The notebook includes functions for reading images, reshaping pixel data, finding optimal clusters, applying segmentation, saving results, and reconstructing images using PCA.

---

📌 Project Overview

This notebook covers:

• Loading and visualizing RGB images
• Reshaping image pixels into a dataset suitable for clustering
• Using the Elbow Method to pick optimal number of clusters
• Applying K-Means to segment images by color similarity
• Reconstructing compressed images using PCA
• Saving segmented images to disk

---

🧭 Workflow Summary

1. Import Required Libraries

Includes Scikit-Learn, NumPy, Matplotlib, Seaborn, PCA modules, and PIL for saving image files.

---

🖼️ Image Loading & Display

Function:

def load_and_show_image(image_path):
    image = mplt.image.imread(image_path)
    ...

This prints image shape, pixel size, and displays the original image.

---

🔄 Reshape Image for Clustering

Images must be reshaped into (num_pixels, 3) before clustering.

Function:

def reshape_image_for_clustering(image):
    X = image.reshape(-1, 3)

This prepares pixel colors (RGB) as dataset rows.

---

📉 Elbow Method to Find Optimal k

Function:

def find_optimal_clusters(X, max_clusters=10):
    ...

The function:

• Runs K-Means for k = 1 to max_clusters
• Stores WCSS (Within-Cluster Sum of Squares)
• Plots an elbow curve to help choose the best value of k

---

🎨 Apply K-Means Image Segmentation

Function:

def apply_kmeans_segmentation(X, image, n_clusters=3):
    ...

This:

• Clusters pixels into k color groups
• Replaces each pixel with its cluster's centroid color
• Reshapes it back to the original image dimensions
• Displays the segmented image

This produces a simplified, abstract version of the image.

---

💾 Save Segmented Image

Function:

def save_segmented_image(image_Seg, output_path='segmented_image.jpg'):
    ...

This stores the segmented/clustering output as a JPEG file.

---

🧠 PCA-Based Image Compression

Function:

def apply_pca(X, image, n_components=2):
    ...

Steps:

• PCA reduces pixel dimensions from 3 → n_components
• The image is reconstructed using inverse transform
• Original and compressed images are displayed side-by-side

This demonstrates how PCA reduces noise and smoothens an image.

---

🖥️ Main Execution Flow

Example:

image_path = 'diver.jpg'
image = load_and_show_image(image_path)
X = reshape_image_for_clustering(image)
find_optimal_clusters(X)
segmented_image = apply_kmeans_segmentation(X, image, n_clusters=3)
save_segmented_image(segmented_image, 'segmented_diver.jpg')
pca_reconstructed_image = apply_pca(X, image, n_components=2)

---

🎯 Goals of the Project

• Understand color clustering using K-Means
• Learn how to reshape image data for ML algorithms
• Apply the Elbow Method to find optimal clusters
• Explore PCA for image compression & reconstruction
• Create reusable modular functions for image processing

---

⚙️ Technologies Used

• Python 3
• NumPy
• Matplotlib
• Scikit-Learn (KMeans, PCA)
• PIL (for saving images)

---

🚀 How to Run This Notebook

1. Install required libraries:

   pip install numpy matplotlib seaborn scikit-learn pillow

2. Place your image (e.g., diver.jpg) in the working directory.

3. Run the notebook sequentially.

4. Adjust n_clusters or n_components for experimentation.

---

✨ Author

This project demonstrates unsupervised learning for image processing using clustering and PCA.

---

If you'd like, I can also:

• Add silhouette score evaluation
• Add comparison with hierarchical clustering
• Generate a combined notebook README portfolio