1: \begin{abstract}

2: The mean-shift algorithm is a popular algorithm in computer vision

3: and image processing. It can also be cast as a minimum

4: gamma-divergence estimation. In this paper we focus on the

5: ``blurring'' mean shift algorithm, which is one version of the

6: mean-shift process that successively blurs the dataset. The analysis

7: of the blurring mean-shift is relatively more complicated compared

8: to the nonblurring version, yet the algorithm convergence and the

9: estimation consistency have not been well studied in the literature.

10: In this paper we prove both the convergence and the consistency of

11: the blurring mean-shift. We also perform simulation studies to

12: compare the efficiency of the blurring and the nonblurring versions

13: of the mean-shift algorithms. Our results show that the blurring

14: mean-shift has more efficiency.  

15: 

16: {\bf keywords}

17: Mean-shift, Convergence, Consistency, Clustering,

18: $\gamma$-divergence, Super robustness.

19: \end{abstract}