Towards Recommender Systems for Police Photo Lineup

摘要

Photo lineups play a significant role in the eyewitness identificationprocess. This method is used to provide evidence in the prosecution andsubsequent conviction of suspects. Unfortunately, there are many cases wherelineups have led to the conviction of an innocent suspect. One of the keyfactors affecting the incorrect identification of a suspect is the lack oflineup fairness, i.e. that the suspect differs significantly from all othercandidates. Although the process of assembling fair lineup is both highlyimportant and time-consuming, only a handful of tools are available to simplifythe task. In this paper, we describe our work towards using recommender systemsfor the photo lineup assembling task. We propose and evaluate two complementarymethods for item-based recommendation: one based on the visual descriptors ofthe deep neural network, the other based on the content-based attributes ofpersons. The initial evaluation made by forensic technicians shows thatalthough results favored visual descriptors over attribute-based similarity,both approaches are functional and highly diverse in terms of recommendedobjects. Thus, future work should involve incorporating both approaches in asingle prediction method, preference learning based on the feedback fromforensic technicians and recommendation of assembled lineups instead of singlecandidates.