Spectral Clustering and Segmentation

We implemented Spectral Clustering which is a popular unsupervised clustering technique suitable for non-convex dataset. We then use this method to segment grayscale and RGB images. To make the clustering more efficient for images, we investigate more into parameter tuning and parameter estimation using minimal spanning trees and K-NN distances, different types of graph Laplacians, sparse matrix representation, upscaling, downscaling and post-processing steps based on Stochastic ensemble consensus that can improve the performance at low-quality cost.

Spectral Clustering	Image Segmentation

Set up:

pip install -r requirements.txt

Please see:

Examples:

Spectral Clustering
Image Segmentation using Spectral Clustering

References:

A Tutorial on Spectral Clustering
Normalized cuts and image segmentation
Stochastic ensemble consensus
Enabling scalable spectral clustering for image segmentation

Authors:

bosakad
harshnehal1996