Linear Dimensionality Reduction and Feature Selection for Visual Object Recognition (pp. 225-241)
Authors: (F. Dornaika, A. Assoum, Department of Computer Science and Artificial Intelligence, University of the Basque Country UPV/EHU, San Sebastian, Spain, and others)
Abstract: Previous work on Linear Dimensionality Reduction (LDR) has emphasized the is- sues of classification and dimension estimation. However, relatively less attention has been given to the critical issue of feature selection in embedded spaces. The main trend in feature extraction has been representing the data in a lower dimensional space, for example, using the Principal Component Analysis (PCA) without using an effective scheme to select an appropriate set of features/eigenvectors in this space. This chapter addresses Linear Dimensionality Reduction through feature selection for visual object recognition. It has two main contributions. First, we propose a unified framework for one transform-based LDR techniques. Second, we propose a framework for two transform-based LDR techniques. As a case study, we consider PCA, Linear Dis- criminant Analysis (LDA), and Locality Preserving Projections (LPP) for the linear methods. We have tested our proposed frameworks on several public benchmark data sets. Experiments on ORL, UMIST, YALE, and PF01 face databases and MNIST handwritten digit database show significant performance improvements in recognition that are based on feature selection in embedded spaces.