robust pca - nus computingleowwk/cs6101/ay2012-13 sem... · 2012. 9. 18. · rpca reconstructed...
TRANSCRIPT
![Page 1: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/1.jpg)
Robust PCA
Yingjun Wu
![Page 2: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/2.jpg)
Preliminary: vector projection
Scalar projection of a onto b:
a1 could be expressed as:
b=(10,4)
w=(1,0)
Example
![Page 3: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/3.jpg)
Preliminary: understanding PCA
![Page 4: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/4.jpg)
Preliminary: methodology in PCA
• Purpose: project a high-dimensional object onto a low-dimensional subspace.
• How-to: – Minimize distance;
– Maximize variation.
![Page 5: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/5.jpg)
Preliminary: math in PCA
• Minimize distance
Energy function
Compress
Recover
![Page 6: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/6.jpg)
Preliminary: PCA example
• Original figure
RGB2GRAY
![Page 7: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/7.jpg)
Preliminary: PCA example
• Do something tricky:
compress decompress
![Page 8: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/8.jpg)
Preliminary: PCA example
• Do something tricky:
compress decompress
Feature#=1900
Feature#=500
Feature#=10 Feature#=50
![Page 9: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/9.jpg)
Preliminary: problem in PCA
PCA fails to account for outliers.
Reason: use least squares estimation.
![Page 10: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/10.jpg)
robust PCA
One version of robust PCA: L. Xu et.al’s work.
Mean idea: regard entire data samples as outliers.
Samples are rejected!
![Page 11: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/11.jpg)
robust PCA
Xu’s work modified the energy function slightly and penalty is added.
If Vi=1 the sample di is taken into consideration, otherwise it is equivalent to discard the sample.
Penalty item
![Page 12: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/12.jpg)
robust PCA
Another version of robust PCA: Gabriel et.al’s work, or called weighted SVD.
Mean idea: do not regard entire sample as outlier. Assign weight to each feature in each sample. Outlier features could be assigned with less weight.
![Page 13: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/13.jpg)
robust PCA
Weighted SVD also modified the energy function slightly.
Original feature Decompressed feature Weight
![Page 14: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/14.jpg)
robust PCA
Flaw of Gabriel’s work: cannot scale to very high dimensional data such as images.
Flaw of Xu’s work: useful information in flawed samples is ignored; least squares projection cannot overcome the problem of outlier.
![Page 15: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/15.jpg)
robust PCA
To handle the problem in the two methods, a new version of robust PCA is proposed.
Still try to modify the energy function of PCA…
Penalty
Distance Scale of error
Outlier process Xu’s work
![Page 16: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/16.jpg)
robust PCA
To handle the problem in the two methods, a new version of robust PCA is proposed.
Still try to modify the energy function of PCA…
Increase without bound! Error rejected!
![Page 17: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/17.jpg)
Experiments Four faces, the second face is contaminated.
Learned basis images.
PCA Xu RPCA
PCA Xu RPCA
Reconstructed faces.
![Page 18: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/18.jpg)
Experiments
Original video
RPCA
PCA
![Page 19: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/19.jpg)
Recent works
• John Wright et.al proposed a new version of RPCA.
• Problem: assume a matrix A is corrupted by error or noise, if we observed D, how to recover A?
Observed matrix
Linear operator
Original matrix error
![Page 20: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/20.jpg)
Recent works
![Page 21: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/21.jpg)
Recent works
![Page 22: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/22.jpg)
Robust PCA demo
![Page 23: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/23.jpg)
References
• De la Torre, F. et.al, Robust principal component analysis for computer vision, ICCV 2001
• M. Black et.al, On the unification of line processes, outlier rejection, and robust statistics with applications in early vision, IJCV 1996
• D. Geiger et.al, The outlier process, IEEE workshop on NNSP, 1991
• L. Xu et.al, Robust principal component analysis by self-organizing rules based on statistical physics approach, IEEE trans. Neural Networks, 1995
• John Wright et. al, Robust Principal Component Analysis: Exact Recovery of Corrupted Low-Rank Matrices by Convex Optimization, NIPS 2009
• Emmanuel Candes et.al, Robust Principal Component Analysis?, Journal of ACM, 2011
![Page 24: Robust PCA - NUS Computingleowwk/cs6101/AY2012-13 Sem... · 2012. 9. 18. · rpca Reconstructed faces. The basis images could be used to reconstruct the original images.\爀伀爀椀最椀渀愀氀](https://reader036.vdocuments.us/reader036/viewer/2022081620/610e27998d54a508854767b4/html5/thumbnails/24.jpg)
Thank you!