Wavelet Clustering Based on Cluster Signal

Wavelet transform can capture the detailed information of signals and discover the fluctuation in the signals. The intention of wavelet clustering is to introduce the wavelet transform to analyze cluster signal and partition the dataset into clusters. Thus, the key is to construct the cluster signal. In this paper, a new wavelet clustering is proposed based on cluster signal. The cluster signal refers to cluster order, which is constructed by neighborhood propagation with the constraint of local density. Cluster order implies the density fluctuation of clusters and wavelet transform can identify the partitioned index in the cluster order and accomplise clustering. Comprehensive experiments show that the proposed wavelet clustering achieve better performance of clustering than compared methods.