Professional Documents
Culture Documents
Muzammilabdulrahman 131103145247 Phpapp02
Muzammilabdulrahman 131103145247 Phpapp02
MUZAMMIL ABDULRAHMAN
2013
Introduction
Application
of FER
Basic Steps of FER
Principal Component Analysis
Local Binary Pattern
Gabor Wavelet Transform
Classification
Simulation Results
Conclusion
References
3
DEFINITION
Disgust
Fear
Happy
Neutral
Sad
Surprise
Automated
Video
access control
surveillance
6
Games
Security
Patient
condition monitoring
The following algorithms can be used in a Holisticbased approach to extract the facial expression
features:
Principal Component Analysis PCA
Linear Discriminant Analysis LDA
Local Binary Patterns LBP
Discrete Wavelet Transform DWT
Gabor Wavelet Transform GWT
Discrete Cosine Transform DCT
PCA
10
The database
a1
a
2
a
2
N
b1
b
2
b
2
N
a1 b1 L h1
r 1 a2 b2 L h2
m
,
M
M M M
a
2 b 2 L h 2
N
N
N
c1
c
2
c
2
N
d1
d
2
d
N 2
where M 213
11
r b2 m2
r a2 m2
r
, c
am
, bm
m
M
M
M
M
aN 2 mN 2
bN 2 m N 2
m1
r d 2 m
c2 m2
2
, dm
,
M
M
M
M
c N 2 mN 2
d N 2 mN 2
e1 m1
r e2 m2
em
,
M
M
eN 2 mN 2
g1 m 1
h1
r h2
g 2 m2
, hm
M
M
M
g N 2 m N 2
hN 2
f1 m1
r f 2 m2
r
, g
fm
m
M
f N 2 mN 2
m1
m2
M
m 2
N
12
r r r r r r r r
A am bm cm d m em f m g m hm
Cov AA
A
A
Find the M eigenvalues and eigenvectors
Eigenvectors of Cov and L are equivalent
U AV r
V is Matrix of
Eigenvectors
r
r r r r r r
A am bm cm d m em f m g m hm
14
A: collection of the
training faces
U: Face Space /
Eigen Space
Compute for each face its projection onto the face space
r
1 U am , 2 U
r
5 U em , 6 U
r
r
bm , 3 U cm , 4 U
r
r
f m , 7 U g m , 8 U
r
dm ,
r
hm
15
rN 2
m
r 2
2
rm
M M
rN 2 mN 2
16
r
U rm
17
Different illumination
18
19
233
224
150
200
173
185
120
128
20
1
T h re s h o ld
1
0
0
1
B in a ry : 1 0 0 0 1 0 1 1
D e c im a l: 1 3 9
20
21
Uniform Pattern: An LBP is called uniform if the binary pattern contains at most
two bitwise transitions from 0 to 1 or vice versa when the bit pattern is
considered circular
Example
The patterns 00000000 (0 transitions)
01110000 (2 transitions)
are uniform
11001111 (2 transitions)
The patterns 11001001 (4 transitions) and 01010011 (6 transitions) are not uniform.
Advantages of Uniform LBP
P
Save memory: With a non uniform pattern there is
Possible combinations while for uniform LBP there are patterns
P( Pof 1) 2
Uniform LBP detects only the important local textures like spots,
edges and corners
22
23
1
x, y, ,
e
2
2
X ' Y ' 2
(
)
2
i X '
2
24
26
28
for i 1.. M
31
Experiment
PCA [1]
SVM [4]
Recognition
80.00
95.11
88.79
85.57
94.5
128x96
230X250
64X64
44X32
Rate (%)
Face Dimension 128x96
40X30
32
33
34
35