Comparative Analysis of Back Propagation Neural Network and Self Organizing Feature Map in Estimating Age Groups Using Facial Features
E. O. Omidiora
Department of Computer Science and Engineering, Faculty of Engineering and Technology, Ladoke Akintola University of Technology, Ogbomoso, Nigeria
M. O. Oladele
Department of Computer Engineering, School of Engineering Technology, Federal Polytechnic, Ede, Nigeria
T. M. Adepoju *
Department of Computer Engineering, School of Engineering Technology, Federal Polytechnic, Ede, Nigeria
A. A. Sobowale
Department of Computer Engineering, School of Engineering Technology, Federal Polytechnic, Ede, Nigeria
O. A. Olatoke
Department of Computer Engineering, School of Engineering Technology, Federal Polytechnic, Ede, Nigeria
*Author to whom correspondence should be addressed.
Abstract
Aim: This paper presents a statistical analysis of the performance of two age estimation algorithms namely Back Propagation Neural Network (BPNN) and Self Organizing Feature Map (SOFM) on human face images.
Methodology: 630 human face images with age ranges 0 - 69 from the FG-NET database were considered, feature extraction was done using Principal Component Analysis (PCA) and classification was done using BPNN and SOFM. Two way ANOVA was used to analyse if there is significant difference between the two algorithms (BPNN and SOFM) by feeding in all the parameters such as training time, number of correctly classified, number of near-correctly classified, number of incorrectly classified and percentage accuracy.
Results: The results of the analysis shows that there is significant difference between BPNN and SOFM in the age estimation using facial features.
Conclusion: The results from the statistical analysis (Analysis of Variance (ANOVA)) reveals that SOFM is better than BPNN because F-critical > F for the column and the decision rule states that we accept H0 i.e. there is significant difference between BPNN and SOFM if F-critical > F when the results (training time, testing time, number of correctly classified, number of incorrectly classified and accuracy) were compared and tested.
Keywords: Age estimation, back propagation neural network, self organizing feature map, principal component analysis, facial features