A bidirectional 2-D linear discriminant analysis algorithm based on an adaptively weighted function
Guochang Gu · Harbin Gongcheng Daxue Xuebao/Journal of Harbin Engineering University · 2008
Two-dimensional linear discriminant analysis extracts feature vectors directly from an image matrix.This improves the speed of feature vector extraction and also eliminates problems posed by small samples.However,the quantity of feature vector data obtained using this method quickly become so large as to make the classification task difficult,and the optimal projective matrix can only be derived from the column direction of the image.Moreover,different samples have different effects on the optimal projective matrix.To solve these weaknesses,we propose a two-dimensional linear discriminant algorithm based on an adaptively weighted function,whereby the image matrix is analyzed bidirectionally.Two-dimensional linear discriminant analysis is done in the horizontal and vertical directions sequentially;the different samples are given different weights so as to improve the performance of classification in the low-dimensional linear space.Numerical experiments on the ORL and Yale facial databases showed that the proposed method outperforms the original two-dimensional linear discriminant analysis algorithm while reducing the feature vector's dimensions.