Kernel Auto-associator from Kernel Principal Component Autoregression with Application to Face Recognition

Based on the kernel principal component regression (KPCR) recently proposed in the literature, a new kernel auto-associator (KAA) model is proposed for classification and novelty detection. For face recognition problem, KAA model can efficiently characterize each subject thus offering a good recognition performance. Steming from the Principal component regression (PCR). a simple technique using principal components as a subset selection method in regression, KPCR selects a subset of the principal components from the kernel space for the response variables to regress. As an extension of KPCR, a kernel auto-associator model can be built from autoregression by first extracting features in the kernel space and then performing ordinary least square reconstruction from the selected features. To demonstrate the performance of the proposed KAA model, face recognition is studied as a benchmark example, with a modular scheme which is consist of two stages. The first stage is a preprocessing by a multilevel two-dimensional (2D) discrete wavelet transform. The second stage is a subject-specific KAA structure, which means each subject is assigned a KAA model for coding the corresponding visual information. Experiments on several well-known face datasets demonstrated high recognition accuracies by just using a few of the kernel principal components