A Novel Face Recognition System Based on Subclass Kernel Nonparametric Discriminant Analysis

摘要

This paper presents a Novel Face Recognition System Based on Subclass Kernel Nonparametric Discriminant Analysis (SKNDA), which incorporates subclass information in the KNDA optimization process. It jointly minimizes the data dispersion within and between subclasses to improve classification accuracy. Moreover, SKNDA has the advantage of reducing data dimension in a discriminative way and performing a discriminant eigenspace representation, where some near-global variations of the data are incorporated in the kernel space, to handle heteroscedastic, non-normal and non-linearly separable face classes. In order to make adequate face recognition, we integrate the Gabor features and the ordinal measures to derive the facial features, which are encoded in local regions, as visual primitives. The different ordinal measures are extracted from Gabor filtering responses. Then, the statistical distributions of these primitives in diverse face image blocks are concatenated, which generates a feature vector whose dimension is reduced using PCA. Finally, the latter is employed as a feature input for the proposed SKNDA. An extensive comparison of the SKNDA model to relevant existing kernel classifiers is performed on real world datasets to show the advantages of our proposed method. In particular, the experiments on face recognition have clearly shown the superiority of the SKNDA over other methods.