Maximum n ormalized spacing for efficient visual clustering

1. Maximum normalized spacing for efficient visual clustering Presenter: Cheng-Hui Chen Author: Zhu-Gang Fan, Yadong Wu, Bo Wu CIKM 2010

2. Outlines • Motivation • Objectives • Methodology • Experiments • Conclusions • Comments

3. Motivation Many of the existing clustering methods often fail to learn the whole structure of themultiple manifolds and they are usually not very effective.

4. Objectives This paper propose a simple distance metric learning method called Maximum Normalized Spacing (MNS)which is a generalized principle based on Maximum Spacing. Combining both the internal and external statistics of clusters to capture the density structure of manifolds.

5. Methodology MST Clustering An extension Establish  MST Normalized spacing Remove the largest edge Determining the number of clusters

6. Methodology Maximizing spacing The generated k clusters have maximum spacing.

7. Methodology

8. Methodology • Min-max cut • Normalized spacing NSP (k)

9. Determining the number of clusters • Via coding length • K is the kernel Gram matric • Distance metric

10. Determining the number of clusters The cluster bisectioning step can be automatically stopped when H > 0. This computing is expensive.

11. Speeding Up for Large Databases

12. Experiments

13. Experiments Image distance metric Clustering accuracy

14. Experiments

15. Experiments

16. Experiments

17. Conclusions Our experimental results show that MNS method is consistently accurate, efficient and it has some advantages over some of the state-of-the-art clustering methods. MNS can be used for many fields of real world.

18. Comments • Advantages • It has some advantages over of the state-of-the-art clustering methods. • Applications • Visual Clustering