Professional Documents
Culture Documents
10/14/2016 1
Why dimensionality reduction?
Some features may be irrelevant
To visualize high dimensional data
“Intrinsic” dimensionality may be smaller than the number of
features
10/14/2016 2
What is PCA?
Why PCA?
10/14/2016 3
PCA Algorithm
PCA algorithm:
X Create N x d data matrix, with one row vector xn per
data point
X subtract mean x from each row vector xn in X
Σ Find covariance matrix of X
Find eigenvectors and eigenvalues of Σ
PC’s the M eigenvectors with largest eigenvalues
10/14/2016 4
Steps in PCA: 1
Calculate Adjusted Data Set
Mi is calculated
… n … -
= by taking the
dims
mean of the
values in
dimension i
data samples
10/14/2016 5
Steps in PCA: 2
Calculate Co-variance matrix, C, from Adjusted Data Set, A
Co-variance Matrix: C
Cij = cov(i,j)
10/14/2016 6
Steps in PCA: 3
Calculate eigenvectors and eigenvalues of C
Matrix E
Matrix E
Eigenvalues Eigenvalues
x x
Eigenvectors Eigenvectors
If some eigenvalues are 0 or very small, we can essentially discard those eigenvalues
and the corresponding eigenvectors, hence reducing the dimensionality of the new
basis.
10/14/2016 7
Principal Components
5 1st principal vector
4
project 1
Minimum RMS 0
error -1
Principal vectors -2
are orthogonal -3
10/14/2016 8
10/14/2016 9
Unitary Matrix
What is Unitary Matrix
A Matrix A is a Unitary Matrix if
A-1 = A*T
A* =Conjugate of A
A*T = Transpose of Conjugate of A
10/14/2016 10
Singular Value Decomposition
The SVD for Square matrices was discovered independently
by Beltrami in 1873 and Jordan in 1874, and extended to
rectangular matrices by Eckart and Young in 1930s.
The SVD of a rectangular matrix A is a decomposition of the
form
A = U S VT OR A = U S VH
Where A is an m x n matrix.
U and V are orthogonal/ Unitary matrices.
S is a diagonal matrix comprised of singular values of A.
10/14/2016 11
SVD - Properties
10/14/2016 12
SVD Properties
V1T
VnT
13
SVD - Properties
1 1 1 0 0
2 2 2 0 0
1 1 1 0 0
x λ1 x
= u1 u2
5 5 5 0 0
λ2
0 0 0 2 2
0 0 0 3 3 v1
0 0 0 1 1
v2
10/14/2016 14
Singular value decomposition
SVD A = U S VT
n n
k R kR
kR
VT
m m U S
A
Y
10/14/2016 15
SVD of a Matrix
= * *
M = U S VT
10/14/2016 16
SVD - Example
A = U S VT - example:
1 1 1 0 0 0.18 0
2 2 2 0 0 0.36 0
1 1 1 0 0 0.18 0 9.64 0
5 5 5 0 0 = x 0 5.29
x
0.90 0
0 0 0 2 2 0 0.53
0 0 0 3 3 0 0.80 0.58 0.58 0.58 0 0
0 0 0 1 1 0 0.27 0 0 0 0.71 0.71
10/14/2016 17
SVD – Dimensionality reduction
1 1 1 0 0 0.18 0
2 2 2 0 0 0.36 0
1 1 1 0 0 0.18 0 9.64 0
5 5 5 0 0 = x 0 5.29
x
0.90 0
0 0 0 2 2 0 0.53
0 0 0 3 3 0 0.80 0.58 0.58 0.58 0 0
0 0 0 1 1 0 0.27 0 0 0 0.71 0.71
10/14/2016 18
SVD - Dimensionality reduction
1 1 1 0 0 0.18
2 2 2 0 0 0.36
1 1 1 0 0 0.18 9.64
5 5 5 0 0 ~ x x
0.90
0 0 0 2 2 0
0 0 0 3 3 0 0.58 0.58 0.58 0 0
0 0 0 1 1 0
10/14/2016 19
SVD - Dimensionality reduction
1 1 1 0 0 1 1 1 0 0
2 2 2 0 0 2 2 2 0 0
1 1 1 0 0 1 1 1 0 0
5 5 5 0 0 ~ 5 5 5 0 0
0 0 0 2 2 0 0 0 0 0
0 0 0 3 3 0 0 0 0 0
0 0 0 1 1 0 0 0 0 0
10/14/2016 20
Applications of SVD in Image Processing
SVD approach can be used in Image Compression.
SVD can be used in face recognition.
SVD can be used in watermarking.
SVD can be used for Texture classification.
SVD can be used for Dynamic Texture Synthesis.
10/14/2016 21
Introduction
10/14/2016 22
Texture Types
Static Texture- as an image showing spatial stationarity.
Dynamic Texture- a sequence of images showing temporal stationarity.
10/14/2016 23
Dynamic Texture Synthesis
10/14/2016 24
Comparison
Between different approaches for Dynamic Texture Synthesis
Different Physics- Image based
Parameters based Patch Parametric
Analysis Cost HIGH LOW LOW
Synthesis Cost HIGH LOW LOW
Model Size LOW HIGH LOW
Specificity HIGH LOW LOW
Flexibility HIGH LOW HIGH
10/14/2016 25
SVD-Soatto-Doretto Model
The temporal mean of each pixel is computed stored in a vector d ЄR m and subtracted from
Then obtained a new matrix by using SVD,
10/14/2016 27
Analysis
The columns of U and V are defined as the left and right singular vectors,
respectively, and S is a diagonal matrix formed by the singular values
of ordered according decreasing energy.
10/14/2016 28
Analysis
The generic column vector yk is thus:
where xk is the k-th column of matrix X and e is the residual error.
Written differently, we see that
10/14/2016 30
Synthesis
Second Step-
The final model is thus represented by the following system of
equations:
zk = Cxk + d.
10/14/2016 31
Survey
I. Standard SVD model by- Soatto and Doretto.
It perform dimension reduction of the data by applying the SVD to the video
frames unfolded into column vectors. This permits only to exploit the
temporal correlation. Drawbacks
10/14/2016 32