Adviser: Bo-Chi Lai

guineanhillΤεχνίτη Νοημοσύνη και Ρομποτική

20 Οκτ 2013 (πριν από 4 χρόνια και 20 μέρες)

103 εμφανίσεις

Speaker: Yi
-
Chun
Ke

Adviser: Bo
-
Chi Lai



Ku
-
Yaw Chang


Introduction


Material and Method


Results


Conclusion

2009/10/06

1


Traditional
Garbo

texture description


two
-
dimensional Gabor function


m(x, y) = |gmn(x, y)


i(x, y)|






μ


mean
δ

standard deviation



s

scale k

orientation


Descriptors=2
×

s
×

k + 2

2009/10/06

2


Rayleigh
Garbo

texture description


1
-
D Gabor function


m(x) = |gmn(x)


i(x)|





s

scale k

orientation


Descriptors=s
×

k + 2



2009/10/06

3


Back propagation neural network(BPNN)

2009/10/06

4


Traditional Gabor texture descriptor


S=4 scales


K=6 orientations


Descriptors=2
×

s
×

k + 2=50


Rayleigh model Gabor texture descriptor


S=4 scales


K=6 orientations


Descriptors= s
×

k + 2=26

2009/10/06

5


Back propagation neural network(BPNN)


input nodes number:50 or 26


output nodes number: 4


hidden nodes:
10



2009/10/06

6



2009/10/06

7

Data 2

Data 1

Data 3

dataset 1

traditional Gabor descriptor


dataset 1

Rayleigh model Gabor descriptor

2009/10/06

8

dataset 2

traditional Gabor descriptor

dataset 2

Rayleigh model Gabor descriptor

2009/10/06

9

dataset 3

traditional Gabor descriptor

dataset 3

Rayleigh model Gabor descriptor

2009/10/06

10

2009/10/06

11



2009/10/06

12

Rayleigh

Traditional

Training time

More

Less

Computational expense

Less

More

Accuracy

Low

High

Lose some performance compared

Easy

Hard


James A. Freeman and David M.
Skapura
,
Netuoral

Networks
Algorithms,Applications,and

Programming
Techniques,1991,90
-
93


Sitaram

Bhagavathy
,
Jelena

Te
si

c, and B. S.
Manjunath
,
On the Rayleigh Nature of Gabor Filter Outputs, Digital
Object Identifier 10.1109/ICIP Volume 3,2005, I11


745
-

I11


748


Xu

Zhan,
Xingbo

Sun, Lei
Yuerong
, Comparison of two
gabor

texture descriptor for texture classification ,
Information Engineering, 2009. ICIE '09. WASE
International Conference on Volume 1, 2009, 52


56


Technology Exponent


http://www.tek271.com/?about=docs/neuralNet/IntoToNeural
Nets.html

2009/10/06

13