The traditional spectral clustering algorithm is sensitive to initialization. Aiming at this defect, we introduce the canopy algorithm to conduct coarse cluster and get the initial clustering center as the input of the K-Means algorithm. Then we propose a spectral clustering algorithm based on canopy clustering (CanopySC) to reduce the blind selection of the initial center of the traditional spectral clustering algorithm. We apply the new algorithm to face image clustering. Compared with the traditional spectral clustering algorithm, the Canopy-SC algorithm can not only get better clustering centers and results, but also has a higher clustering accuracy. Experiments demonstrate its effectiveness and feasibility.

Key words: