Genetic & Evolutionary Biometrics: Hybrid Feature Selection and Weighting for a Multi-Modal Biometric System

The Genetic & Evolutionary Computation (GEC) research community is seeing the emergence of a new and exciting subarea, referred to as Genetic & Evolutionary Biometrics (GEB), as GECs are increasingly being applied to a variety of biometric problems. In this paper, we present successful GEB techniques for multi-biometric fusion and multi-biometric feature selection and weighting. The first technique, known as GEF (Genetic & Evolutionary Fusion), seeks to optimize weights for score-level fusion. The second technique is known as GEFeWSML (Genetic & Evolutionary Feature Weighting and Selection-Machine Learning). The goal of GEFeWSML is to evolve feature masks (FMs) that achieve high recognition accuracy, use a low percentage of features, and generalize well to unseen subjects. GEFeWSML differs from the other GEB techniques for feature selection and weighting in that it incorporates cross validation in an effort to evolve FMs that generalize well to unseen subjects.