Information Technology Reference
In-Depth Information
Table 7.1
Comparison of training and testing performance for ORL face dataset with different
feature extraction techniques and different neuron architectures
S. No. Neuron type Network Parameters Feature
Average FRR FAR
Recognition
type
extraction epochs
rate (%)
×
1
R
MLP
48-8-1
40
401
R
PCA
44,000
0.092 0.032
96.6
48-7-1
40
×
351
R
ICA
6,000
0.067 0.014
98.25
R
PCA
28,000
0.083 0.020
97.8
2
C
MLP
48-3-1
40
×
151
C
PCA
28,000
0.092 0.018
98.0
R
ICA
6,000
0.087 0.016
98.2
C
ICA
6,000
0.067 0.0095 98.9
R
PCA
28,000
0.080 0.018
98.0
3
C
RPN
48-2-1
40
×
101
C
PCA
28,000
0.080 0.016
98.2
(
d
=
0
.
9)
R
ICA
6,000
0.075 0.018
98.0
C
ICA
6,000
0.070 0.011
98.75
R
PCA
6,000
0.092 0.020
97.7
4
C
RSS
48-1
40
×
99
C
PCA
6,000
0.080 0.014
98.3
R
ICA
4,000
0.083 0.017
98.1
C
ICA
4,000
0.075 0.009
99.0
R
PCA
6,000
0.086 0.020
97.85
5
C
RSP
48-1
40
×
99
C
PCA
6,000
0.075 0.009
98.8
R
ICA
4,000
0.083 0.013
98.4
C
ICA
4,000
0.070 0.006
99.25
Fig. 7.6
Recognition rate
versus subspace dimension
for different feature extraction
techniques in ORL face
database
In order to assess the sensitivity of different feature extraction methods on the
number of subspace dimension variations, the comparative performance is shown
in Fig.
7.6
. The results of experiments are typical in part, because the recognition
rate does not increases monotonically with the number of subspace dimensions.
Search WWH ::
Custom Search