Bridging the gap in recognition memory between unknown and known faces

Lead Research Organisation: University of Stirling
Department Name: Psychology

Abstract

Recognizing a known face is usually effortless, but most of us experience uncertainty with new acquaintances - is that the person I saw yesterday? Research has shown that such 'unknown' face processing is surprisingly poor and that it is difficult to match a picture, perhaps on a passport, to a person. Error rates of 20% are common. This has serious implications for our security, since it is not possible to be sure of the identity of people entering the country, for example. The aim of this project is to study the process of becoming familiar with a face. Doing so will yield insights into how our memories work, and why they fail, and help to establish the best training to improve performance of those who need to match faces, such as border guards.

An obvious difference between a known face and an unknown one is the amount of experience we have of the face. We interact with friends on a regular basis, seeing them in different situations and from various angles. It seems that this enables us to create a memory for their face, which allows us to recognize it again in other situations, perhaps years later. Previous work has begun to study how we learn new faces, by presenting a variety of images; we propose to draw on the best of this work and, crucially, to study what is happening inside the brain during learning.

Our method of observing brain function uses a measure of the electrical fields generated by the brain called Event-Related Potentials (ERPs). Previous work has suggested that there are two separate processes underlying our memory: familiarity and recollection. Familiarity, in this context, is the all too common experience - I've seen that face before, but where? Recollection brings relief, as you remember who it is and why you know them. Familiarity and recollection produce distinct patterns of electrical activity, which will allow us to monitor the process of developing memory and establish the best ways to encourage it.

We plan a series of experiments to study the process of learning a face, starting with simple repetition and progressing to varied views, methods to encourage thoughtful consideration of each face, and the effects of sleep on helping consolidate memories of new faces. The aim is to assess whether memory performance for unknown and known faces is associated with the same retrieval processes, or whether different retrieval processes are involved. Establishing the involvement of different retrieval processes in memory for known and unknown faces is theoretically important because it would link the poor performance in face recognition for unknown faces to differences in the mental representation underlying memory for different types of face. Clarifying the reason why known and unknown face recognition differs also has important implications for how face recognition can best be improved. Having identified the retrieval processes engaged, various mnemonic strategies will be used to promote face learning and 'bridge the gap' in recognition memory accuracy between the two types of face. Importantly, the analysis of brain activity will reveal exactly how changing the way unknown faces are studied leads to better memory - either by enhancing the deployment of retrieval processes associated with unknown faces, or by actually changing the processes engaged to match those used for known faces. Answering this question will provide important insight into the nature of memory representations that support successful face recognition, with implications for theoretical accounts of both episodic memory and face identification - informing future attempts to improve face recognition in real world contexts.

Technical Summary

Recognition memory is better for known faces than for unknown faces, possibly due to existing semantic representations of known faces providing retrieval cues that are unavailable for unknown faces. This project will investigate whether encoding manipulations can improve recognition memory for unknown faces. This project will investigate the interaction between semantic representations and episodic retrieval in the context of recognition memory for faces using event-related potential (ERP) measures of brain activity. Recognition memory studies using ERPs have identified neural correlates of retrieval processes termed familiarity and recollection; while familiarity provides strength-based quantitative information recollection can provide qualitative information about the context in which an item was encoded. Unknown faces seem to be recollected differently from both words and known faces, two types of stimulus that have pre-existing semantic representations. Our hypothesis is that a lack of pre-existing semantic memory representations leads to a decrease in memory performance that is associated with activity of atypical retrieval processes. This project will first identify neural correlates of familiarity and recollection for both known and unknown faces, which has not been done before, and will then begin manipulating encoding conditions across experiments in an attempt to improve memory performance. Frequency of encoding, exposure to exemplar density, elaborative encoding and sleep-dependent consolidation will be manipulated. The translation of findings from basic to applied research will be achieved by using the same manipulations in forensically-inspired face identification tests. In this manner, the project will both contribute to theory concerning the interaction between semantic memory and episodic retrieval, and provide useful guidance on ways to improve memory for unknown faces that may impact positively on face recognition in applied settings.

Planned Impact

Two sectors will benefit from the project findings aside from beneficiaries in the immediate academic field. The principal beneficiary is the UK public sector, including the Home Office and several subsidiary and allied departments. In addition, project outcomes are likely to interest and engage the general public, who are everyday stakeholders in face recognition and its failings and who think face recognition is important due to its social relevance. Vehicles to deliver impact will be a stakeholder workshop, public engagement activities and creation of a website to publicize the project that will incorporate a face recognition training module.

Home Office: The project will identify ways to improve recognition memory for unknown faces that can inform practice in government bodies such as the Border Force and Police. In addition, we will deliver scientific evidence concerning the use of memory strategies that can promote successful recognition of target individuals. Global security requires the detection of terror suspects and project outcomes will provide useful guidance on improving the chances of recognizing a suspect at airports and other border controls. Furthermore, project findings will be useful for informing approaches to policing, and have the potential to improve the reliability and efficiency of identifying criminal suspects from identity parades and CCTV footage. Research outcomes may be useful not only for training police officers but for guidance on how eyewitnesses to crimes are instructed in order to maximize the chances of successful criminal detection.

Workshop: a workshop for stakeholders will be held at the University of Stirling at the beginning of the third year of the project. Talks will be given by members of the project team, invited academic partners and government stakeholders (to be contacted by Marek Rejman-Greene from the Centre for Applied Science and Technology at the Home Office). The workshop will provide the opportunity for a wide range of parties to raise issues of concern in face recognition research and the applied face recognition problem, with the aim of disseminating cutting edge knowledge and inspiring future research collaborations between government bodies and universities.

Public engagement: Face recognition has intuitive appeal to the general public and provides popular exhibits at museums and science fairs. The project background and findings will be communicated at the Glasgow Science Museum and Edinburgh's annual Science Festival. The use of known faces in particular has high appeal to the public, and demonstration of some experimental tasks is a fun and informative way to engage audiences.

Web site: A project website describing the background to the applied face recognition problem along with details of the current research project will be launched during the first year of the programme. Stakeholders will be invited to comment on the project, opening the possibility that stakeholders can influence aspects of the project such as the stimuli or experimental design. As the project unfolds, findings will be added to the website, along with details of the stakeholder workshop. Publicity materials will be uploaded in order that they can reach the widest possible number of stakeholders. In addition to describing project information, the website will include a 20-minute 'how to improve your face recognition' training module that will allow people to test their memory for faces at baseline, interpret their results, undertake face recognition training derived from project findings before retesting their face recognition.
 
Description The project investigated human face recognition, which is particularly poor for unknown faces, and represents a significant practical problem for security and police identification. Across a series of experiments using human behaviour and brain imaging we investigated how performance could be improved - to bring it closer to the very good levels of recognition that people achieve for known faces. Our work examined the effects of classic memory encoding strategies (e.g., depth of processing, repetition of the face) and manipulations based on theoretical understanding of face processing (e.g., provision of multiple viewpoints of the face) or memory consolidation (e.g., provision of sleep between encoding and retrieval). Across all of the manipulations there was no strong evidence that memory for unknown faces could be significantly enhanced - some manipulations had no real effect (e.g., depth of processing and sleep), where others did improve performance (e.g., repetition and multiple views) but not to the level of recognition achieved for known faces. Whilst the outcomes of the studies were not what we hoped for, they are helpful in ruling out these kinds of approaches in future attempts to improve face identification behaviour. From a theoretical perspective, we have essentially employed conceptual approaches to encouraging higher level processing mechanisms; the negative results suggest that future research should focus on more perceptually driven manipulations (e.g., visual noise, method of revealing the stimuli). Similarly, although the findings did not provide useful strategies for practitioners in the field to be able to adapt methods used in practice, the end users that we engaged with were interested to discover that approaches that would be expected to work did not do so. Importantly, as well as using laboratory-based neuro-imaging studies, we also tested some of the manipulations using purely behavioural face identification procedures that were modelled on tests used in security and witness identification settings - using line ups of multiple faces. We found equivalent results using these real-world tests as we had in our neuro-imaging studies, suggesting that the outcomes generalise to the kinds of procedures used by practitioners in applied settings.
Exploitation Route Our findings constrain the range of techniques that practitioners should consider using when they attempt to improve face identification in the field. There are classic memory manipulations that might be expected to be worth using as aids, but which have no actual benefit.
Sectors Government, Democracy and Justice,Security and Diplomacy
URL http://recognitionmemory.org
 
Description The project investigated human face recognition, which is particularly poor for unknown faces, and represents a significant practical problem for security and police identification. Across a series of experiments using human behaviour and brain imaging we investigated how performance could be improved - to bring it closer to the very good levels of recognition that people achieve for known faces. Our work examined the effects of classic memory encoding strategies (e.g., depth of processing, repetition of the face) and manipulations based on theoretical understanding of face processing (e.g., provision of multiple viewpoints of the face) or memory consolidation (e.g., provision of sleep between encoding and retrieval). Across all of the manipulations there was no strong evidence that memory for unknown faces could be significantly enhanced - some manipulations had no real effect (e.g., depth of processing and sleep), where others did improve performance (e.g., repetition and multiple views) but not to the level of recognition achieved for known faces. Whilst the outcomes of the studies were not what we hoped for, they are helpful in ruling out these kinds of approaches in future attempts to improve face identification behaviour. From a theoretical perspective, we have essentially employed conceptual approaches to encouraging higher level processing mechanisms; the negative results suggest that future research should focus on more perceptually driven manipulations (e.g., visual noise, method of revealing the stimuli). Similarly, although the findings did not provide useful strategies for practitioners in the field to be able to adapt methods used in practice, the end users that we engaged with were interested to discover that approaches that would be expected to work did not do so. Importantly, as well as using laboratory-based neuro-imaging studies, we also tested some of the manipulations using purely behavioural face identification procedures that were modelled on tests used in security and witness identification settings - using line ups of multiple faces. We found equivalent results using these real-world tests as we had in our neuro-imaging studies, suggesting that the outcomes generalise to the kinds of procedures used by practitioners in applied settings. As promised in our Pathways to Impact plan, our grant involved interaction with Marek Rejman-Greene, the Senior Biometrics Advisor at the Centre for Applied Science and Technology (CAST) at the Home Office. We had a series of discussions with Marek about the work, and towards the end of the grant period he attended a workshop at Stirling where we presented the outcomes of our studies, along with a series of other academics (presenting findings on similar face-recognition studies). Marek expressed how important this work was in efforts to enhance face identification and security operations, and we discussed a future collaboration - he asked us if we would be able to write a review of the current state of the art of approaches to manipulating face identification, written for an applied audience. Subsequently, however, Marek retired from his role at the Home Office, and the work did not go ahead. The work was highlighted at a number of other public events, including a talk at a Science Festival, and a Science Cabaret event, and via a dedicated website for the project. These dissemination activities were successful, raising clear interest in the public and helping educate lay audiences about the significant problem facing police and security forces when they attempt to identify unfamiliar faces.
First Year Of Impact 2017
Sector Government, Democracy and Justice,Security and Diplomacy
Impact Types Cultural