Enhancing CCTV: Averages improve face identification from poor‐quality images

Abstract

Low-quality images are problematic for face identification, for example, when the police identify faces from CCTV images. Here, we test whether face averages, comprising multiple poor-quality images, can improve both human and computer recognition. We created averages from multiple pixelated or nonpixelated images and compared accuracy using these images and exemplars. To provide a broad assessment of the potential benefits of this method, we tested human observers (n = 88; Experiment 1), and also computer recognition, using a smartphone application (Experiment 2) and a commercial one-to-many face recognition system used in forensic settings (Experiment 3). The third experiment used large image databases of 900 ambient images and 7,980 passport images. In all three experiments, we found a substantial increase in performance by averaging multiple pixelated images of a person's face.These results have implications for forensic settings in which faces are identified from poor-quality images, such as CCTV.

Face averages and multiple images in a live matching task

“…Another recent article in which averages from 10 images per identity were created also failed to find an average advantage in a one‐to‐one matching task using high‐quality images. For pixelated images, however, creating averages of these did improve overall performance (Ritchie et al ., ). Therefore, the null effect reported in the current study is not unprecedented in the literature.…”

“…Another study demonstrated that images picked out by the models themselves produced lower accuracy on a face matching task than images picked out by people unfamiliar with the models (White, Burton, & Kemp, ), highlighting that our impressions of how we look do not align with the impressions of strangers. With familiar people, however, it seems that we can cope with a large range of variability, and increasing familiarity correlates with increasing likeness ratings for variable images of the same person (Ritchie et al ., ). Put simply, once familiar with someone's face, any image of them is judged to be a better likeness, compared with unfamiliar raters.…”

“…In addition to providing multiple images, another candidate for improved matching performance is an average image comprising multiple photographs of the same person. Averages have been shown to produce more accurate computer face recognition than single images (Burton, Jenkins, Hancock, & White, ; Jenkins & Burton, ; Robertson, Kramer, & Burton, ), even when the averages are derived from pixelated images (Ritchie et al ., ). Face averages have also been shown to help with human face recognition when the averages comprise several ambient images (White, Burton, et al., ) and when facial composites are averaged together (Bruce, Ness, Hancock, Newman, & Rarity, ; Hasel & Wells, ).…”

“…Evidence suggests that people may form an average as an internal representation when shown an array of images of a new identity (Kramer, Ritchie, & Burton, ). However, this reported improvement in matching with face averages was not replicated in other work, where no overall benefit was found in human performance (Ritchie et al ., ).…”

Searching for faces in crowd chokepoint videos

“…These averages are thought to provide a stable face representation by diluting idiosyncratic aspects of particular instances (Jenkins & Burton, ), therefore providing greater identity information. Although evidence from face matching tasks is mixed with regard to whether averages improve performance (Ritchie et al, , ; White, Burton, et al, ), the potential benefits of their use during crowd searching have yet to be investigated.…”