Deriving score-based Likelihood Ratios from facial images of different quality: A practical approach

Abstract

In this work, a method for computing the score-based Likelihood Ratio (SLR) in the context of forensic face recognition is presented. The quality of facial images is first assessed through the Open-Source Facial Image Quality (OFIQ) library, which is available on the GitHub platform [1]. The generation of Between-Source Variability (BSV) and Within-Source Variability (WSV) curves for each quality range is achieved by employing two distinct facial image datasets. A generic approach is adopted to facilitate SLR computations across different quality levels, with the aim of enhancing reliability in forensic applications. The proposed method has been thoroughly validated, demonstrating its effectiveness in addressing the challenges posed by varying image quality in forensic scenarios, as well as its practical applicability in disaster victim identification (DVI) situations.