In this article, the clustering output results using Spectral clustering (with normalized Laplacian) are going to be compared with taht obtained using KMeans clustering on a few shape datasets.
The following couple of slides taken from the Coursera Course: Mining Massive Datasets by Stanford University
describe the basic concepts behind the spectral clustering and the spectral partitioning algorithms.
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.
The following figure shows the result of spectral partitioning for automatic separation of foreground from the background on yet another image.