ON EFFICIENCY OF FACE RECOGNITION USING LINEAR DISCRIMINANT ANALYSIS AND PRINCIPAL COMPONENT ANALYSIS

The solution of the face recognition problem by means of principal component analysis (PCA) and linear discriminant analysis (LDA) is being considered. The main idea of this approach is that firstly, we project the face image from the original vector space to a face subspace via PCA, secondly, we use LDA to obtain a linear classifier. In the paper, the efficiency of the PCA+LDA approach to face recognition without preliminary processing (scaling, rotation, translating) is investigated. Research shows that the higher the number of images in a class of teach sample, the higher the face recognition rate. When the number of images is small, face recognition performance can be improved by expanding the training set using the images received by scaling and rotating of initial images. The efficiency of PCA+LDA approach is investigated on the images of ORL database. When processing large sets of images, methods of linear condensation and principal component synthesis are suggested to calculate the main components. The principal component synthesis method is based on splitting an initial image set into small sets of images, obtaining eigen-vectors of these sets (particular solutions) and calculation of eigenvectors of an initial image based on particular solutions. The linear condensation method is based on the decrease of an order of matrix allowing to calculate pretty exactly eigenvectors whose eigenvalues are located in the preset interval. It is shown that linear condensation and principal component synthesis methods allow to decrease significantly the processing time of building a classifier by PCA+ LDA approach, without reducing face recognition rate.