Machine learning approaches for person identification and verification

Abstract

New machine learning strategies are proposed for person identificationwhich can be used in several biometric modalities such as frictionridges, handwriting, signatures and speech. The biometric or forensicperformance task answers the question of whether or not a sample belongsto a known person. Two different learning paradigms are discussed:person-independent (or general learning) and person-dependent (orperson-specific learning). In the first paradigm, learning is from ageneral population of ensemble of pairs, each of which is labelled asbeing from the same person or from different persons- the learningprocess determines the range of variations for given persons and betweendifferent persons. In the second paradigm the identity of a person islearnt when presented with multiple known samples of that person- wherethe variation and similarities within a particular person are learnt.The person-specific learning strategy is seen to perform better thangeneral learning (5% higher performace with signatures). Improvement ofperson-specific performance with increasing number of samples is alsoobserved.