A Comparative Study of Image Segmentation Methods

This project tests a variety of different image segmentation methods by performing image segmentation on a dataset of single-object images. We score the results of each method based on the foreground pixels.

Contents

• Project Report
• Presentation Video
• Dataset: Oxford-IIIT pet dataset
• Segmentation Methods
  • Image Segmenation Model
    • Unet Model
    • Model Predictions
    • Model's Evaluation based on Foreground Pixels of Predicted Mask
  • Otsu Method
  • Edge-Based Segmentation
  • Simple Threshold Method
  • Region Adjacency Graph
• Project Demonstraction
  • Interactive Jupyter Notebook