Multi-Criteria in Discriminant Analysis to Find the Dominant Features

Abstract

A crucial problem in biometrics is enormous dimensionality. It will have an impact on the costs involved. Therefore, the feature extraction plays a significant role in biometrics computational. In this research, a novel approach to extract the features is proposed for facial image recognition. Four criteria of the Discriminant Analysis have been modeled to find the dominant features. For each criterion is an objective function, it was derived to obtain the optimum values. The optimum values can be solved by using generalized the Eigenvalue problem associated to the largest Eigenvalue. The modeling results were employed to recognize the facial image by the multi-criteria projection to the original data. The training sets were also processed by using the Eigenface projection to avoid the singularity problem cases. The similarity measurements were performed by using four different methods, i.e. Euclidian Distance, Manhattan, Chebyshev, and Canberra. Feature extraction and analysis results using multi-criteria have shown better results than the other appearance method, i.e. Eigenface (PCA), Fisherface (Linear Discriminant Analysis or LDA), Laplacianfaces (Locality Preserving Projection or LPP), and Orthogonal Laplacianfaces (Orthogonal Locality Preserving Projection or O-LPP).