Biometric Verification by Projections in Error Subspaces

Abstract

A general methodology for design of biometric verification system is presented. It is based on linear feature discrimination using sequential compositions of several types of feature vector transformations: data centering , orthogonal projection onto linear subspace, vector component scaling, and orthogonal projection onto unit sphere. Projections refer to subspaces in global, within-class, and between-class error spaces. Twelve basic discrimination schemes are identified by compositions of subspace projections interleaved by scaling operations and single projection onto unit sphere. For the proposed discriminant features, the Euclidean norm of difference between query and average personal feature vectors is compared with the threshold corresponding to the required false acceptance rate. Moreover, the aggregation by geometric mean of distances in two schemes leads to better verification results. The methodology is tested and illustrated for the verification system based on facial 2D images.