🖼️ Image Processing Script

✨ Features

This script utilizes several advanced techniques for precise image segmentation and artistic processing:

• Edge Detection - Identify object boundaries
• Morphological Operations - Refine object masks
• K-means Clustering - Separate objects from backgrounds
• Object Extraction - Isolate and process objects independently
• Comic Style Processing - Apply artistic effects to foreground objects (posterization)
• Background Modification - Apply Gaussian blur to background
• Object Detection & Classification - Identify specific objects using machine learning (KNN)
• Performance Evaluation - Calculate IoU accuracy with ground truth (Segmentation)

🚀 Setup Instructions

Clone the Repository

git clone https://github.com/usergaia/IMAGE-PROC-PROJECT.git
cd IMAGE-PROC-PROJECT

Dependencies

• MATLAB or VSCode with MATLAB Extension
• Statistics and Machine Learning Toolbox (for K-means clustering and KNN classification)
• Image Processing Toolbox (for advanced morphological operations)

File Structure

IMAGE-PROC-PROJECT/
├── images/
│   ├── raw/               # Place input images here
│   ├── ground_truth/      # Ground truth images for accuracy evaluation
│   └── ground_truth_mask/ # Ground truth masks
├── models/                # Stores segmentation models
└── src/                   # Contains processing scripts
└── main.m

Note: To run the gt_mask_converter, ensure that this is in the same hierarchy as the images folder.

📋 Execution Workflow

Step 1: Read and Resize Image

• The input image is read from the 'images/raw/' directory
• Ground truth images and masks are loaded for later accuracy evaluation
• All images are resized to 512×512 pixels for standardization

Step 2: Apply Canny Edge Detection and Morphological Gradient

• Convert the image to grayscale for edge detection
• Apply Canny edge detection
• Enhance edges using a morphological gradient

Step 3: Create Initial Mask and Reduce Noise with Morphological Operations

• Fill holes in the edges to generate an initial mask
• Refine using morphological operations:
  • Apply closing to connect nearby edges
  • Apply opening to remove small noise
  • Remove small disconnected regions (<500 pixels)
• Keep only the largest connected component if multiple objects are detected

Step 4: Apply K-means Only on the Masked Region with the Color Space LAB

• Convert image to LAB color space for better color-based segmentation
• Extract pixels within the initial mask for clustering
• Apply K-means clustering (k=2) only on the masked region
• Determine the main cluster (typically the larger one) to represent the object
• Create a K-means based mask by selecting pixels from the main cluster

Step 5: Refine K-means Segmentation with Morphological Operations

• Apply closing to connect nearby parts
• Fill any remaining holes in the mask
• Remove small objects (<500 pixels)
• Keep only the largest connected component in the final mask

Step 6: Extract Objects Using Final Mask

• Extract foreground object by multiplying the image with the final mask
• Extract background by multiplying the image with the inverse of the final mask

Extra: Modify the Extracted Objects

Foreground - Comic Style Processing

• Apply posterization (8 levels) for cartoon-like appearance
• Boost colors by a factor of 5 to enhance vibrancy
• Increase saturation by 40% in HSV color space
• Add a warm tone to the foreground with factors [1.1, 1.0, 0.85]
• Create black outlines using Canny edge detection and dilation
• Apply the comic style only to the masked foreground area

Background - Gaussian Blur

• Apply Gaussian blur (sigma=10) to the background for depth effect

Combine Foreground and Background

• Start with the blurred background
• Add the comic-styled foreground on top

Step 7: Display Results

The script creates a visualization showing:

• Original Image
• Each intermediate step of the segmentation process (8 steps)
• Processed background and foreground components
• Final combined output with styling

Step 8: Object Detection on Segmented Object in the Image

• Extract region properties (BoundingBox) from the final mask
• Load a pre-trained KNN model from 'models/knnModel.mat'
• Extract image features using the extractImageFeatures function:
  • Calculate texture features using Gray Level Co-occurrence Matrix (GLCM)
  • Extract Haralick features (energy, contrast, homogeneity, entropy)
  • Compute normalized color histograms for each RGB channel
• Predict the object class from 5 categories:
  • Burj Khalifa
  • Basketball
  • Car
  • Clown Fish
  • Logitech Mouse
• Draw bounding boxes and labels on the final image

Step 9: Performance Evaluation (Segmentation IoU)

• Preprocess masks for comparison
• Ensure both predicted (final_mask) and ground truth masks are binary
• Calculate Intersection over Union (IoU):

  IoU = (Intersection / Union) * 100
  

• Display IoU accuracy percentage
• Create a visual comparison showing:
  • Expected extracted object (ground truth)
  • Actual extracted object
  • Binary ground truth mask
  • Binary predicted mask

📊 Segmentation Accuracy Evaluation

The script automatically evaluates segmentation accuracy using:

• Intersection over Union (IoU) - A standard metric for segmentation quality
• Visual comparison between expected and actual results
• Console output showing exact IoU percentage