Finding and visualizing relevant subspaces for clustering high-dimensional astronomical data using connected morphological opera