Facial Feature Extraction Method Based on Coefficients of Variances

Principal Component Analysis (PCA) and Linear Discriminant Analysis(LDA) are two popular feature extraction techniques in statisticalpattern recognition field. Due to small sample size problem LDA cannotbe directly applied to appearance-based face recognition tasks. As aconsequence, a lot of LDA-based facial feature extraction techniques areproposed to deal with the problem one after the other. Nullspace Methodis one of the most effective methods among them. The Nullspace Methodtries to find a set of discriminant vectors which maximize thebetween-class scatter in the null space of the within-class scattermatrix. The calculation of its discriminant vectors will involveperforming singular value decomposition on a high-dimensional matrix. Itis generally memory- and time-consuming.Borrowing the key idea in Nullspace method and the concept ofcoefficient of variance in statistical analysis we present a novelfacial feature extraction method, i.e., Discriminant based onCoefficient of Variance (DCV) in this paper. Experimental resultsperformed on the FERET and AR face image databases demonstrate that DCVis a promising technique in comparison with Eigenfaces, NullspaceMethod, and other state-of-the-art facial feature extraction methods.