Machine Recognition in Complex Domain - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 155

Fig. 7.10 Graph of variance
captured by each principal
component of Indian face
database

Table 7.2 Comparison of training and testing performance for Indian face dataset with different
feature extraction techniques and different neuron architectures
S.no. Neuron Network Parameters Feature Average FRR FAR Recognition

type extraction epochs rate (%)

1 R MLP 60-8-1 50 × 497 R PCA 45,000 0.19 0.055 93.8

R ICA 25,000 0.17 0.045 95.3

R PCA 30,000 0.20 0.044 95.0

2 C MLP 60-3-1 50 × 187 C PCA 30,000 0.21 0.046 94.8

R ICA 10,000 0.17 0.040 95.5

C ICA 10,000 0.17 0.035 96.6

R PCA 30,000 0.19 0.047 94.8

×

3 C RPN 60-2-1 50 125 C PCA 30,000 0.20 0.047 94.8

( d = 0 . 94) R ICA 10,000 0.19 0.045 95.2

C ICA 10,000 0.18 0.037 96.1

R PCA 20,000 0.17 0.042 95.3

4 C RSS 60-1 50 × 123 C PCA 20,000 0.18 0.046 95.0

R ICA 10,000 0.20 0.041 95.5

C ICA 10,000 0.16 0.029 96.8

R PCA 20,000 0.15 0.041 95.4

5 C RSP 60-1 50 × 123 C PCA 20,000 0.16 0.043 95.2

R ICA 10,000 0.20 0.037 95.9

C ICA 10,000 0.17 0.025 97.2

From extensive simulations on this face dataset, it is imperative to make following
inferences:

•

Class distinctiveness is very poor in case of R PCA and C PCA .