Comparing Spectral partitioning / clustering (with Normalized Graph Laplacian) with KMeans Clustering in R

The following simpler spectral partitioning approach (thresholding on the second dominant eigenvector) can also be applied for automatic separation of the foreground from the background.