Comparing Spectral clustering (with Normalized Graph Laplacian) with KMeans Clustering

In this article, the clustering output results of Spectral clustering (with normalized Laplacian) is going to be compared with KMeans on a few shape datasets.

1. The following algorithm (by Ng. et al) is going to be used for spectral clustering (with normalized Gaussian).

2. The following figures / animations show the spectral clustering results with different Gaussian similarity kernel bandwidth parameter values on different shape datasets and also the comparison with the results obtained using the KMeans counterpart.

3. Finally, both KMeans and Spectral clustering (with bandwidth=0.1) algorithms are applied on an apples and oranges image to segment the image into 2 clusters. As can be seen from the next image, the spectral clustering could separate out the orange while the KMeans could not.