Using comparative human descriptions for soft biometrics

Abstract

Soft biometrics is a new form of biometric identification which utilizes labeled physical or behavioral traits. Although these traits intuitively have less discriminatory capability than mensurate approaches, they offer several advantages over traditional biometric techniques. Soft biometric traits can be typically described as labels and measurements which can be understood by people, allowing retrieval and recognition based solely on human descriptions. Although being a key component of eyewitness evidence, conventional human descriptions can be considered to be unreliable. A novel method of obtaining human descriptions will be introduced which utilizes visual comparisons between subjects. The Elo rating system is used to infer relative measurements of subjects' traits based on the comparative human descriptions. This innovative approach to obtaining human descriptions has been shown to counter many problems associated with categorical (absolute) labels. The resulting soft biometric signatures have been demonstrated to be robust and allow accurate retrieval of subjects in video data and show that elapsed time can have little effect on comparative descriptions.