Clustering by differencing potential of data field

Hierarchical clustering with data field can find clusters with various shape and filter the noises in data set without input parameters. However, its clustering process is complex and cannot effectively deal with complex and high dimensional data. In this paper, a novel clustering algorithm is proposed by differencing potential (DP) of data field. The potential difference specifies the nearest object which has high potential as the aggregation direction, and the data distance is used to divide the global data set into local multiple clusters. Simultaneously, noises are identified effectively in the light of the potential of data field. Experimental results on eight popular data sets and a facial image data set indicate that the proposed method outperforms existing clustering algorithms for dealing with data set with high dimensions and distribution in complex shape, as well as noise identification.