An Investigation on Recent Advances in Feature Partitioning Based Principal Component Analysis Methods

Principal Component Analysis (PCA) is one of the well-known and widely accepted dimensionality reduction techniques in varied domains. However, PCA does not scale well computationally with increasing dimensionality and it extracts only global features, ignoring local features. The local features may be very useful for classification. More recently, partitioning based PCA approaches (FP-PCA) have been proposed to compute principal components to overcome these shortcomings. In this paper we analyze the existing FPPCA methods and classify them into meaningful categories for better understanding. Subsequently we bring out the properties of these FP-PCA methods. This analysis provides the basis for further research on the FP-PCA approaches which appear to promise improvements in classification as well as savings in computation. We also show the superiority of these recent FPPCA methods using our experimentation on ORL and Yale face data sets.