Face Recognition at its best: Leveraging science and policing for innovative human-machine solutions

Superior Face Processing
Face processing

To bring together police practitioners and scientists interested in face identity processing, I am organizing a symposium at the European Police Congress (May 11-12, 2022). For more detailed symposium information expand this post.

Meike Ramon
20 Apr 2022

[Want a German version? Drop me an email! Click here to get to the website of the European Police Congress.]

Symposium summary:

Face Recognition at its best: Leveraging science and policing for innovative human-machine solutions

Processing facial identity is a critical component of everyday police work - from standard identification verification from ID cards to targeted search and surveillance for persons of interest. Traditionally, face identity processing occurred in direct interactions. Today, however, as the volume of image and video material increases, processing of digital traces represents an increasingly complex challenge for police agencies worldwide. This symposium presents a unique collection of talks highlighting findings from police practitioners working closely with research scientists. Their efforts aim to provide the evidence and information that is needed to address police agencies challenges today - and anticipate those of the future. Six talks presented by 10 speakers from agencies and universities in Germany, Switzerland, and the USA will summarize state-of-the-art applied research that is critically relevant for human-based face identity processing and machine-based face recognition in policing.

1) Perpetrator identification from “wild” CCTV footage - improvements through Super-Recognizers? René Rolvien & Maren Mayer

Over the past years, there has been an increasing interest in so-called “Super-Recognizers” – individuals with above average face processing capacities that have been identified using lab-tests of face processing. Despite considerable knowledge gaps, there is an increasing demand for application of Super-Recognizers in law enforcement. Indeed, many agencies are interested in identifying them, have been proffered assessment via tests developed for research purposes. However, whether Super-Recognizers identified with lab-tests represent can offer an actual benefit in real-life policing settings remains unestablished. Here we present the first study to formally address this question. In a practitioner-initiated study, we assessed perpetrator identification performance using real case material from criminal investigations in controls and individuals identified as SRs based on a novel lab-based diagnostic framework (Ramon, 2021).

2) Evaluating lab-based tests for assessment of face processing skills for police purposes Michael Vomland & Jeffrey Nador

Numerous studies of facial identity processing (cl)aim to measure skills that are critical for real-life demands of security professionals. Typically they employ tests developed by researchers, using procedures originating from the domains of experimental psychology and face processing. None of these studies have, however, investigated whether/which solutions substantially capture professionally relevant abilities. Therefore, combining expertise from policing, work and organizational psychology, as well as human face cognition, we investigated the relationship between police officers’ (N=114) ability in two work samples and their performance on challenging lab-based tests of face perception and recognition. Analyses reveal that while officers’ behavior on work samples is highly correlated, behavior in these work samples is captured relatively poorly by standard lab tests. These results demonstrate that even highly challenging tests of face processing used to identify “Super-Recognizers” in the lab may not fully capture professionally relevant abilities, and imply that more bespoke testing could improve this process.

3) Real-world Super-Recognizer identification in the Berlin Police using beSureⓇ Meike Ramon & Simon Rjosk

Face recognition "in the wild" remains the most challenging task for algorithms and humans alike. Since face identity processing is untrainable but also crucial across officers' everyday tasks, police agencies are interested in identifying so-called "Super-Recognizers" (SRs) to improve their operations. Unfortunately, existing solutions for SR selection have two major shortcomings: they do not capture police officers' real-world tasks, nor do they involve the type of image material characteristic of authentic police work. We combined expertise from science and policing to overcome these limitations and identify SRs among the Berlin Police's 25K employees. We present the Berlin Test for SR Identification and our vision for establising a common standard in SR work by sharing our tools with interested agencies worldwide.

4) Forensic use of face recognition systems for investigation Maëlig Jacquet

With the increasing development of automatic systems and artificial intelligence, face recognition is becoming increasingly important in forensic and civil contexts. However, face recognition has yet to be thoroughly empirically studied to provide an adequate scientific and legal framework for investigative and court purposes. This observation sets the foundation for the research. We focus on issues related to face images and the use of automatic systems. Our objective is to validate a likelihood ratio computation methodology for interpreting comparison scores from automatic face recognition systems (score-based likelihood ratio, SLR). We collected three types of traces: portraits (ID), video surveillance footage recorded by ATM and by a wide-angle camera (CCTV). The performance of two automatic face recognition systems is compared: the commercial IDEMIA Morphoface (MFE) system and the open source FaceNet algorithm.

5) Identification of persons of interest visible in images: review of the current process in police and areas for improvement Melina Vona & Olivier Delemont

The proliferation of images (CCTV, smartphones, …) capturing persons of interest in relation to criminal activities (before, during or after the perpetration of an offense) has created a new kind of traceability of such kind of activities. Faced with this new field of investigation, police services have developed strategies aimed at making the best use of the information that can be derived from these images, either with a view to recognizing the person of interet or to linking offences committed by the same (group of) perpetrators. These strategies are more often than not grounded in a pragmatic approach, inherently derived from the operation and structure of a given police service. This may raise questions about the relevance and effectiveness of these approaches, especially in these times of great hype about the promises of AI. The purpose of this talk is to share the findings of a study that assessed the procedure implemented in the police Neuchâteloise, a small to medium sized state police force in Switzerland. This procedure relies on the distribution of a request for recognition/identification by e-mail to all employees of this state police. This request contains the image of the person of interest, and summary information about the case. Success rates and response times of this procedure were recorded, and a survey was carried out to collect the views and opinions of the police officers. In a second stage, images of a person of interest for which an identification was formally confirmed were submitted to the Better Tomorrow facial recognition software, and the number of times the software returned the correct person in its list of candidates was computed. It was shown that the success rate of this facial recognition solution was highly dependent on the quality of the information transmitted by the image. For images in which the face of the person of interest was well detailed, the facial recognition solution provided very good results in a short time. But as soon as the quality of the facial information decreases, the identification rate of the recognition procedure by sending to police officers became greater.

6) Detecting Deepfakes - comparing humans and machines Matt Groh

The recent emergence of machine-manipulated media raises an important societal question: How can we know whether a video that we watch is real or fake? In two online studies with 15,016 participants, we present authentic videos and deepfakes and ask participants to identify which is which. We compare the performance of ordinary human observers with the leading computer vision deepfake detection model and find them similarly accurate, while making different kinds of mistakes. Together, participants with access to the model’s prediction are more accurate than either alone, but inaccurate model predictions often decrease participants’ accuracy. To probe the relative strengths and weaknesses of humans and machines as detectors of deepfakes, we examine human and machine performance across video-level features, and we evaluate the impact of preregistered randomized interventions on deepfake detection. We find that manipulations designed to disrupt visual processing of faces hinder human participants’ performance while mostly not affecting the model’s performance, suggesting a role for specialized cognitive capacities in explaining human deepfake detection performance.

Speaker Information (alphabetic order):

  • Olivier Delemont, Professor, Ecole des sciences criminelles, University of Lausanne (CH) Prof. Dr. Olivier Delémont likes to describe himself as a generalist in forensic science. After obtaining a degree (1996) and a PhD (2005) in Forensic science, he pursued his academic career for about ten years, while working in parallel at the Forensic science unit of the Geneva Police(CH). Since then, he has devoted himself entirely to his work at the Ecole des sciences criminelles (School of Criminal Justice) of the University of Lausanne. His research activities and interests cover a wide range of forensic science areas, in particular those related to crime scene investigation, and to the use of traces in police investigations.
  • Matt Groh, PhD Candidate, MIT Media Lab (USA) Matt Groh is a PhD student at the MIT Media Lab in the Affective Computing lab (USA). He studies the influence of collective and artificial intelligence on individual decision-making. This involves conceptualizing, prototyping, and evaluating data-driven sociotechnical systems spanning deepfake detection, teledermatology diagnosis, automated emotion recognition, and more. Before MIT, Matt cofounded Proprio Labs and worked as a data scientist at two startups (Qadium, Raise Me), two research organizations (the World Bank and DARPA) and a non-profit (Innovations for Poverty Action). Matt has a master’s degree in Media Arts and Sciences from MIT and a bachelor’s degree from Middlebury College where he majored in economics and minored in Arabic and mathematics.
  • Maëlig Jacquet, Postdoc, Ecole des sciences criminelles, University of Lausanne (CH) After obtaining her B.Sc. degree in Biology at the University of Cergy-Pontoise in France, Dr. Maëlig Jacquet moved to Switzerland to study forensic science at the Ecole des Sciences Criminelles (School of Criminal Science) of the University of Lausanne. She received her M.Sc. in Forensic Identification in 2015, and her Ph.D. thesis dedicated to forensic face recognition in 2021. Her research mainly focuses on the use of face recognition systems for investigation and court purposes as well as on the interpretation of forensic evidence. Through her work, she emphasises the importance of connection between the academic researches and the operational needs. Dr. Jacquet is a member of the Digital Imaging Working Group (DIWG) of the European Network of Forensic Science Institutes (ENFSI).
  • Maren Mayer, PhD Candidate, University of Mannheim (DE) Maren Mayer is a PhD student at the University of Mannheim and the Heidelberg Academy of Sciences and Humanities. She studies collaborative behavior and group decision making focusing on expert influences. Her research involves experimental examination as well as statistical modelling of collaborative behavior disentangeling various sources of influence on the group judgment. Maren has a bachelors's and a master's degree in psychology obtained at the University of Mannheim and is currently an associated member of the research training group "Statistical Modeling in Psychology".
  • Jeffrey Nador, Postdoc, Applied Face Cognition Lab (CH) Originally from Montreal, Quebec, after completing his bachelor’s in Honors Psychology at Concordia University (CA), Jeff received his master’s and PhD in Psychology from Northeastern University (USA), followed by postdoctoral positions at Wright State University (USA) and the University of Fribourg (CH). He has worked collaboratively with the Sridhar Neurotechnology Group and Office of Naval Research in the United States, as well as the Rhineland-Palatinate Police (DE). His current research employs data-driven analyses of neural and behavioral data sets to examine influences of attention and memory, in terms of how these factors affect performance.
  • Meike Ramon, Assistant Professor, Applied Face Cognition Lab (CH) Prof. Dr. Meike Ramon is a German cognitive neuroscientist and internationally renowned expert on face identity processing. Funded by the Swiss National Science Foundation to investigate the Mechanisms of Superior Face Recognition she founded and heads the Applied Face Cognition Lab at the University of Lausanne (previously hosted at the University of Fribourg). She advises international governments, policy makers, and law enforcement on matters concerning human and automatic face recognition, and maintains active collaborations with international police agencies. Since 2017, she is an Advisor to the Berlin Police and Berlin State Office of Criminal Investigation.
  • Simon Rjosk, Detective, Innovation & Science Manager, Center for Innovation and Science Management, State Office of Criminal Investigation Berlin (DE) Detective Simon Rjosk (M.Sc.) is the Innovation and Science Manager of the State Office of Criminal Investigation in Berlin and among other things responsible for the identification of Super-Recognizers within the Berlin Police. Rjosk joined the Berlin Police in 2011 as a trained social scientist, graduated as a detective and worked his way up through various stations, e.g. narcotics investigation and crime scene investigation, until he joined the strategic innovation management in 2017. Since then, he has made it his mission to further develop the Berlin Police in a knowledge-based manner and make it more future-proof. As part of his master's thesis in science management, which he completed extra-professionally, he developed a strategic concept for the establishment of a Center for Innovation and Science Management. Based on his thesis, this unique department was created, in which Rjosk himself has since been working, to meet the increasingly complex fight against crime with knowledge. Because as a "Police-Science-Hybrid" Rjosk is convinced that security authorities will only be able to tackle future challenges by collaborating with researchers and experts, to utilize their knowledge and make it an integral part of their identity.
  • René Rolvien, Detective Inspector Candidate, Berlin Police; BA student, Berlin School of Economics and Law (DE) René Rolvien is a police detective trainee at the Berlin School of Economics and Law. Before coming to the Berlin Police, he had studied American Studies at the Johannes Gutenberg University Mainz, obtaining his Bachelor's and Master's degree there. He is also interested in psychology that is why he decided to write his Bachelor thesis about Super-Recognizers in the context of police work to finish his studies at the Berlin School of Economics and Law and to eventually become a police detective. Simlutaneously, he is a student of psychology at the FernUniversität Hagen which is a state distance-learning university.
  • Michael Vomland, Head of the Neuwied Criminal Police Investigation Department, Neuwied Criminal Police & Rhineland-Palatinate Police University (DE) Michael Vomland is the head of a CID at the Koblenz police headquarters. Various units of his CID are daily engaged in human facial recognition in different ways. As part of a thesis at the German Police University in Münster, M.V. has combined the connections between academic research findings on facial recognition and police practice. At present, M.V. is working together with scientists on the implementation of persons with outstanding face processing abilities into the practice of the police in Rhineland-Palatinate.
  • Melina Vona, Forensic Scientist, Department of Forensic Science & Crime Intelligence, Neuchatel Police; Ecole des sciences criminelles, University of Lausanne (CH) Melina Vona is a forensic scientist at the state police of Neuchâtel, Switzerland. She obtained a bachelor and master degrees in forensic science at the Ecole des sciences criminelles (School of Criminal Justice) of the University of Lausanne. Her master’s thesis was performed in collaboration with the state police of Neuchâtel and concerned the identification of persons visible in various images (e.g. from surveillance camera or taken by witnesses). The main aim was to estimate the ability of police officers, and of a facial recognition software, to recognise people in images.