Biometric Classification with Factor Analysis

Abstract

This research presents a study on biometrics classification using Factor Analysis (FA). As a multivariate statistical tool, factor analysis is useful for understanding the underlying structure in a dataset. Moreover, in addition to achieving an economy of the variables, the

"factors" or hypothetical constructs can provide an alternate yet succinct representation of the data. It is a method of determining, from an observable set of variables, a basic set of components that are common to all the observations. In this study, the loadings (or

weights) on the Factors are used to classify the data in alternate representation. In particular, we will examine and group the data according to three biometric features. In the first part, we demonstrate the capabilities of factor analysis to capture the gender of the individual. This will enable us to use FA as a gender classifier. The next study will show the use of an FA as a facial hair classifier. Given a group of individuals, we will be able to classify them as either having beards or not. Finally, in the last part presented in

this work, we will work on classifying the facial expressions of a group of Japanese women. Given all seven universal expressions per subject (two or three of each expression), we will use factor analysis to group each subject according to their expression. Furthermore, given an individual with a particular expression, we will use factor analysis as a biometric measure in the determination of the particular expression exhibited.