Algorithm 1.
Step 1: | Construct the graph G on the given data set X and calculate the similarity matrix W; |
Step 2: | Generate the normalized Laplacian matrix L̃ = D–1/2LD–1/2; |
Step 3: | Obtain the K relaxed continuous solutions (the first K eigenvectors) of Eq. (4) based on the eigenvalue decomposition on L̃; |
Step 4: | Generate the final discrete solution via K-means [18] or spectral rotation [6], [10] based on these continuous solutions. |