The direction of this work is in the field of subspace face recognition, where faces are assumed to belong to a face space, a subspace of the original high-dimensional space. The primary goal of this work is to present a qualitative and quantitative comparison of the three most representative subspace analysis algorithms in face recognition, namely: Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Independent Component Analysis (ICA), and at the same time gain the necessary experience to apply and enhance these methods in the future, to achieve better recognition results. The motivation of this task is the lack of concrete comparison between the three algorithms in the face recognition literature.; Results show that the PCA and ICA appear to be stable as the number of training images is increased, while the classification results of LDA become noticeably lower than the two former algorithms. The classification rates do not appear to be affected by the variation of subspace dimensions in the FERET experiments. (Abstract shortened by UMI.)
展开▼
