Study on multi-center fuzzy C-means algorithm based on transitive closure and spectral clustering.

Highlights: [•] The multi-center initialization method can get rid of the problems that the FCM algorithm is sensitive to the initial prototypes, and it can handle non-traditional curved clusters. [•] The similarity matrix generated by the Floyd algorithm is notably block symmetric, which ensure the effective extraction of spectral features. [•] The problem of clustering samples is transformed to a problem of merging subclusters, the computational load is low, and has strong robustness. [Copyright &y& Elsevier]