For DBSCAN algorithm can't cluster some variable density data sets effectively, VDBSCAN solved this question by a k-dist figure to automatically obtain the neighborhood radiuses of various density levels of data objects. k-VDBSCAN algorithm obtains the parameter k automatically to reduce the parameter k ‘sinfluence in the final clustering results. Based on the data set itself, using the characteristics of the distance between the data objects, on the basic of the k-VDBSCAN algorithm, an clustering algorithm based on improved parameter k obtained automatically is proposed. Theoretical analysis and experimental results show that the improved algorithm can effectively automatically obtain the value of the parameter k and the clustering results and time efficiency has improved significantly.