A Genetic Programming-PCA Hybrid Face Recognition Algorithm173
Table 1. Examples of Acquired Relationship Functions for
Detecting Image Group 1. PCA[n] Is the Nth Value on PCA
Vector.
No Function
1 (PCA[8]- MAX(PCA[19], PCA[7]))> PCA[12]
2 AND((PCA[2]< PCA[13]), MIN(PCA[1], 1))
3 (PCA[0]× NOT(PCA[5]))
4 (PCA[1]× (PCA[21]> (PCA[2]-PCA[18])))
few components and as a result have a relatively low
computational overhead.
Results are brought in Table 2, where they are com-
pared to EigenFace [3] and SVM [13] clustering methods.
It is observed that Genetic Programming without lever-
aging has the worst results. On the other hand, Leveraged
Genetic programming beats other methods in Five-to-
Five. In leave-one-out the results are repeated for Genetic
Programming, although this time Leveraged Genetic
Programming fell %2.5 (one image in total of 40 images)
short of SVM.
While it could be inferred that the Genetic Program-
ming is usable as a feature detector, it is believed that
PCA is limited in reducing data dimension [14]. Further
research is required to investigate Genetic Program-
ming’s results with 2D PCA and Multilinear PCA.
To further investigate Genetic Programming’s per-
formance, number of partitioned class groups was
changed and the results were brought in Table 3. It was
observed that the further partitioning of the images in-
creases recognition error, while decreasing k might man-
dates increase in time spent for Genetic Programming’s
evolution.
6. Conclusions
Genetic programming is a general purpose search algo-
rithm that can be utilized in classification problems. In
this paper, Genetic programming was exploited to clas-
sify face images. The results showed that Genetic Pro-
gramming alone is not suitable, as required time and
computational overhead surpasses that of other methods,
and also its recognition ratio is usually lower.
Table 2. Comparision of Different Algorithms Recognition
Rate.
Method Five-to-Five Leave One Out
Eigenface 87.0% 85.0%
SVM 91.0% 95.0%
GP 63.5% 67.5%
Leveraged GP 91.5% 92.5%
Table 3. Effect of Number of Partitions in Leveraged Ge-
netic Programming on Recognition Rate.
Number of Partitions Recognition Rate
2 88.5%
4 91.5%
5 91.5%
8 91.0%
10 89.0%
To improve results, a leveraging algorithm was ap-
plied to Genetic Programming. The leveraged Genetic
Programming showed a good recognition rate, compara-
ble to or in some cases even better than that of other me-
thods.
REFERENCES
[1] S. Liu, Y. Tian and D. Li, “New research Advances of
Facial Expression Recognition,” International Conference
on Machine Learning and Cybernetics, Baoding, Vol. 2,
July 2009, pp. 1150-1155.
[2] I. T. Jolliffe, “Principal Component Analysis,” Springer-
Verlag New York, Inc., 2002.
[3] M. Turk and A. Pentland, “Eigenfaces for Recognition,”
Journal of Cognitive Neurosicence, Vol. 3, No. 1, 1991, pp.
71-86.
[4] A. Pentland, B. Moghaddam and T. Starner, “View-Based
and Modular Eigenspaces for Face Recognition,” Pro-
ceedings CVPR’94, 1994 IEEE Computer Society Con-
ference on, Seattle, July 1994, pp. 84-91.
[5] A. Eleyan and H. Demirel, “PCA and LDA Based Face
Recognition Using Feedforward Neural Network Classi-
fier,” Lecture Notes in C omputer Scien ce, Vol. 4105, 2006,
pp. 199-206.
doi:10.1007/11848035_28
[6] J. R. Koza, “Genetic Programming: On the Programming
of Computer by Means of Natural Selection,” MIT Press:
Cambridge, 1992.
[7] S. Xuesong and Y. Zhou, “Gray Intensity Images Proc-
essing for PD Pattern Recognition Based on Genetic Pro-
gramming,” International Joint Conference on Artificial
Intelligence JCAI’09, Haikou, 2009, pp. 711-714.
[8] A. Teredesai and V. Govindaraju, “Issues in Evolving GP
Based Classifiers for a Pattern Recognition Task,” Pro-
ceedings of the 2004 IEEE Congress on Evolutionary
Computation, 20-23 June 2004, pp. 509-515.
[9] J. R. Koza, M. A. Keane, M. J. Streeter, W. Mydlowec, J.
Yu and G. Lanza, “Genetic Programming IV: Routine
Human-Competitive Machine Intelligence,” Kluwer Aca-
demic Publishers, Norwell, 2003.
[10] N. Krause and Y. Singer, “Leveraging the Margin More
Carefully,” Proceedings of the Twenty-First International
Conference on Machine Learning, Banff, 2004, p. 63.
Copyright © 2011 SciRes. JSIP