Upload
abner-terence-cain
View
223
Download
0
Embed Size (px)
Citation preview
Intelligent Database Systems Lab
國立雲林科技大學National Yunlin University of Science and Technology
A Novel Density-Based Clustering Framework by Using Level Set Method
Presenter : Zhen-Feng Weng
Authors : Xiao-Feng Wang and De-Shuang Huang
2010/02/23
TKDE.17 (2009)
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
2
Outline
Motivation Objective Method Experiments Conclusion Comments
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
3
Motivation
DBSCAN is very sensitive to the selection of MinPts and Eps. The drawbacks of other approaches:
The overfitting phenomenon. The confusion between cluster boundary and noises.
Clusters
DBSCAN
Real
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
4
Objective
It proposed a level set method for clustering. Base on the assumption that the cluster centers can be
regarded as the target objects (image segmentation).
The cluster center:To find a local maximum of density function.
The image segmentation:To find a local optimum of the intensity function.
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
5
Concept
It applies Geometric Active Contour (GAC) model to obtain the boundaries of clusters. Elastic Evolution
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
6
Overview
Nonparametric density estimation
Initialization of Level Set Evolution
Level Set Evolution
Level Set Density
Valley seeking
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
7
Kernel Density Estimation & Initial Boundaries
Nonparametric density estimation:
Initialization of LSE:
Gaussian kernel
Δf(x)>0, x is inside BΔf(x)=0, x belongs BΔf(x)<0, x is outside B
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
8
Level Set Evolution
The initial boundaries can be further divided into three types: Single-cluster, Multiple-cluster, No-cluster
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
9
Valley Seeking
It is a graph-based theoretic clustering method. Each tree represents a cluster. According to the previous
step:
Construct the forest:
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
10
Experiments
Comparisons with DBSCAN
DBSCANLevel Set Method
N.Y.U.S.T.
I. M.
Intelligent Database Systems Lab
12
Conclusions
It proposed a novel density-based clustering framework using the level set method. avoid the overfitting phenomenon. solve the confusion problem of cluster boundary
points and outliers