A frequentist estimation of duplicate probability as a baseline for person identification from image and video material using anthropometric measurements

Abstract

Video and image material is becoming increasingly ubiquitous thus its potential as evidence in forensic investigations is growing. Once faces are hidden however, the value of surveillance footage is restricted unless there is another biometric trait that can be observed by camera such as linear body measurements. There is much biological evidence for human body proportions exhibiting much individual variation. Nevertheless, the probability of there being two individuals that match in their respective proportions ultimately determines its usability for the assignment, exclusion and even identification of persons in the forensic domain. This work is concerned with approaches for duplicate probability estimations derived from anthropometric measures. Here, we present a novel frequentist estimation using a dataset of 340 individuals and their respective anthropometric measurements. Drawing on density kernel estimations of measurement dissimilarity, we propose the duplicate probability to be in the order of 10^−15 to 10^−8.