Image Segmentation

The goal is to represent the input image as a fully connected and non-directional graph, and partitioned them into multiple sets of segments that share some common characteristics, such as color or intensity.

The following techniques were implemented:

  • Spectral Clustering (k clusters). [Algorithm]
  • Normalized cuts (k clusters). [Algorithm using the k smallest eigenvalues in Step 3]
  • Recursive Normalized cuts (unspecified clusters). [Algorithm using Ncut(A, B) metric]

Georgios Balaouras
Georgios Balaouras
Data Scientist

My research interests include Computer Vision with a healthy pinch of problem solving.