Combined Central and Subspace Clustering for Computer Vision Application

1
Combined Central and Subspace Clustering for
Computer Vision Application

• Le Lu, Rene Vidal
• John Hopkins University
• (????)

2
Introduction

• Central Clustering
• ?????????????????
• Application
• Image segmentation,
• K-means,EM
• Subspace Clustering
• ???????????
• Application
• Motion segmentation, face clustering with varying
  illumination, temporal video segmentation
• K-subspace, Generalized PCA

3
GPCA?K-means
YZ?????
K-means?B1??? A1??????
GPCA?Y?????, YZ???,??
XY?????
4
K-means
YZ?????
XY?????
K-means????????? ??????????????
5
????

• ???
• ????
• ????????
• ????
• ??

6

• Central Clustering
• K-means
• ????????????
• ?????????????????,?????????????????.
• ????????????????.

• Subspace Clustering
• ,
• K-subspace
• Subspace????
• Subspace?????????,????????????????.
• ?????Subspace????.

Subspace???????
7
?????????????
???xi?1???????????.
????????????
8
Algorithm
GPCA
9

• Computing the membership
• Computing the cluster centers
• ? ??????, ????? ????
• Computing the normal vectors
• ????

10
???????????????
?????
11
Experiments (Simulated Data)

• 3???????,600?
• Subspace?2?,????3?????
• ???????100?(????, sµ1.5)
• Subspace?2090
• ??????3????????????2.5sµ 5sµ
• sb?Noise
• 100?,??

12
Experiments (Simulated Data)
KK
KM
KK
MP
KM
GK
GK
MP
JC
JC

• KM K-mean ?6??????? 2????????
• MP MPPC (Mixture of probabilistic PCA )? 6???????
  2????????
• KK K-subspace??????Subspace?K-means
• GK GPCA??????Subspace?K-means
• JC ????

13
Experiments (Illumination)

• 4 subjects (10 subject???)
• 4 poses 64 illuminations
• 240 320 pixels

14

• GPCAK-means
• Subject5?Subject6?????Subject5????????

15
(No Transcript)
16
Experiments (Video)

• Video sequence ? several video shots
• Each video contains 4 shots