Face Recognition Based on Adaptively Weighted Gabor-LDA

A novelty method of face recognition based on adaptively weighted Gabor linear discriminant analysis (Gabor-LDA) is presented. First, the coefficients of Gabor wavelet transform deriving from a face image are taken as eigenvectors. And then an improved algorithm of principal component analysis (PCA) is proposed. This improved PCA is used to decrease the dimension of the eigenvector. A weight is given to each vector according to the distance between the eigenvector and the others in the same class. The new class means are calculated by the weighted of eigenvectors. The within-class scatter matrix and between-class scatter are reconstructed through the new class means, then the LDA discriminate function is improved. The problem of the class mean of training samples deviates from the center of this class in small samples size case is resolved by this improved LDA discriminate function. The experiment shows that a higher recognition result is obtained in the Yale face databases.