Non - Iterative 3D Linear Discriminant Analysis And Its Application In Face Recognition

With the improvement of science and technology and the development of the society, as a result of the limitation of two-dimensional image information, the 2D face recognition technology recognition rate is very difficult to improve significantly, it can not meet our requirement for facial recognition technology in terms of accuracy. However, compared with the two-dimensional face image, the 3D face images with original face surface geometry information is closer to the objective facts. The traditional approach of face recognition for three-dimensional data is draw the three dimensional space data according to a certain dimension for a super high dimensional data matrix, and then the new matrix data of Two-dimensional face recognition methods, and even on the matrix of the data again after stretching for one-dimensional vector data, and then linear discriminant analysis(LDA) and principal component analysis(PCA), But to do this can obviously destroy the original data structure, and cause the super high dimensional "dimension disaster", seriously affect the operation efficiency of face recognition and identification.This article embarks from the 2D face database, reduction by 2D Gabor wavelet transform face more facial feature information, the 2D face to extract facial feature information into 3D face database. Combined with tensor of thoughts, this paper proposes a three-dimensional linear discriminant analysis(3DLDA) dimension reduction method. The method respectively according to the three dimensional array of three dimensions, in turn, separately on each dimension linear discriminant analysis dimension reduction. The article gets an array of dimension reduction after a nearest neighbor classifier for face recognition. At first, this article will make the threedimensional linear discriminant analysis method and the traditional one dimensional and two dimensional recognition method apply in ORL and Yale, PIE and FERET face database, and the United States postal USPS handwritten data sets, the three dimensional linear discriminant analysis method of recognition rate is higher than the traditional one dimensional and two dimensional method. Due to large 3D face data dimension, even through the three dimensional linear discriminant analysis dimension reduction after the array dimension may yet be bigger, so this paper presents a two phase 3D linear discriminant analysis and linear discriminant analysis(3DLDA + LDA) combinations, in another word,the original high dimension of three dimensional array will again reduce the dimension after a three dimensional linear discriminant analysis dimension reduction through linear discriminant analysis dimension reduction, it means the original data for the two dimension reduction, overcome the "dimension disaster", to improve the recognition rate of face recognition. Also by ORL and Yale, PIE and FERET face database, and the United States postal USPS handwritten dataset is verified. And we find the combination of 3DLDA + LDA method recognition rate is more effective than the 3DLDA method.