Charlie D. Frowd

A forensically valid comparison of facial composite systems

An evaluation of E-FIT, PROfit, Sketch, Photofit and EvoFIT composite construction techniques was carried out in a “forensically friendly format”: composites of unfamiliar targets were constructed from memory following a 3–4-hour delay using a Cognitive Interview and experienced operators. The main dependent variable was spontaneous naming and overall performance was low (10% average naming rate). E-FITs were named better than all techniques except PROfit, though E-FIT was superior to PROfit when the target was more distinctive. E-FIT, PROfit and Sketch were similar overall in a composite sorting task, but Sketch emerged best for more average-looking targets. Photofit performed poorly, as did EvoFIT, an experimental system. Overall, facial distinctiveness was found to be an important factor for composite naming.

The relative importance of external and internal features of facial composites

Three experiments are reported that compare the quality of external with internal regions within a set of facial composites using two matching-type tasks. Composites are constructed with the aim of triggering recognition from people familiar with the targets, and past research suggests internal face features dominate representations of familiar faces in memory. However the experiments reported here show that the internal regions of composites are very poorly matched against the faces they purport to represent, while external feature regions alone were matched almost as well as complete composites. In Experiments 1 and 2 the composites used were constructed by participant-witnesses who were unfamiliar with the targets and therefore were predicted to demonstrate a bias towards the external parts of a face. In Experiment 3 we compared witnesses who were familiar or unfamiliar with the target items, but for both groups the external features were much better reproduced in the composites, suggesting it is the process of composite construction itself which is responsible for the poverty of the internal features. Practical implications of these results are discussed.

An application of caricature : How to improve the recognition of facial composites

Facial caricatures exaggerate the distinctive features of a face and may elevate the recognition of a familiar face. We investigate whether the recognition of facial composites, or pictures of criminal faces, could be similarly enhanced. In this study, participants first estimated the degree of caricature necessary to make composites most identifiable. Contrary to expectation, an anticaricature was found to be best, presumably as this tended to reduce the appearance of errors. In support of this explanation, more positive caricature estimates were assigned to morphed composites: representations that tend to contain less overall error. In addition, anticaricaturing reduced identification for morphed composites but enhanced identification for individual composites. Although such improvements were too small to be of value to law enforcement, a sizeable naming benefit was observed when presenting a range of caricature states, which appeared to capitalize on individual differences in the internal representation of familiar faces.

Parallel approaches to composite production: interfaces that behave contrary to expectation

This paper examines two facial composite systems that present multiple faces during construction to more closely resemble natural face processing. A ‘parallel’ version of PRO-fit was evaluated, which presents facial features in sets of six or twelve, and EvoFIT, a system in development, which contains a holistic face model and an evolutionary interface. The PRO-fit parallel interface turned out not to be quite as good as the ‘serial’ version as it appeared to interfere with holistic face processing. Composites from EvoFIT were named almost three times better than PRO-fit, but a benefit emerged under feature encoding, suggesting that recall has a greater role for EvoFIT than was previously thought. In general, an advantage was found for feature encoding, replicating a previous finding in this area, and also for a novel ‘holistic’ interview.

An evaluation of US systems for facial composite production

Witness and victims of serious crime are normally requested to construct a facial composite of a suspects face. While modern systems for constructing composites have been evaluated extensively in the UK, this is not the case in the US. In the current work, two popular computerized systems in the US, FACES and Identikit 2000, were evaluated against a ‘reference’ system, PRO-fit, where performance is established. In experiment 1, witnesses constructed a composite with both PRO-fit and FACES using a realistic procedure. The resulting composites were very poorly named, but the PRO-fit emerged best in ‘cued’ naming and two supplementary measures: composite sorting; and likeness ratings. In experiment 2, PRO-fit was compared with Identikit 2000, a sketch-like feature system. Spontaneous naming was again very poor, but both cued naming and sorting suggested that the systems were similar. The results support previous findings that modern systems do not produce identifiable composites.

Catching Even More Offenders with EvoFIT Facial Composites

Facial composites are an investigative tool used by police to identify suspects of crime. Unfortunately, traditional methods to construct the face have rather low success rates. We have been developing a new recognition-based method called EvoFIT that requires eyewitnesses to select whole faces from arrays of alternatives. Both published laboratory research and existing police field-trials have found that EvoFIT produces images that are more identifiable than images from traditional systems. In the current paper, we present an evaluation of a more recent version of EvoFIT: in 2010, EvoFIT was deployed in 35 criminal investigations by Humberside police and these images directly led to identification of 21 suspects, equating to 60% success - quadruple the performance of the previous system used within the force. The evaluation also showed that identification of a suspect led to conviction in 29% of investigations (6 out of 21). Overall, a conviction occurred in 17% of cases involving use of an EvoFIT (6 out of 35). We also outline more recent developments which indicate that an arrest is now likely in three out of every four cases in which EvoFIT is used, and a conviction rate of one in five.

Giving Crime the 'evo': Catching Criminals Using EvoFIT Facial Composites

Facial composites are traditionally made by witnesses and victims describing and selecting parts of criminals’ faces, but this method is hard to do and has been shown to be generally ineffective. We have been working on an alternative system called EvoFIT for the past 12 years. Much of the development for this system has been carried out in the laboratory but, more recently, police forces have been formally evaluating it in criminal investigations. The current paper describes three of these police audits. It was found that EvoFIT composites made from real eyewitnesses led to an overall arrest in 23.4% of cases, which is similar to a 24.5% correct naming level found by laboratory research. System performance therefore appears to be effective both using simulated testing procedures and in the hands of the intended user.

Evolutionary generation of faces

Publisher Summary This chapter describes a system that allows the evolutionary generation of near-photographic quality face images. The major anticipated use for such a system would be, allowing eyewitnesses to a crime to construct a likeness of the suspect. Current methods for doing this center on photofit-like composite systems. The user is presented with a large range of possible alternatives for major features, such as nose, mouth, and eyes and attempts to reconstruct a face piecemeal. The original photofit does not work very well, probably because it breaks the face into components in a way that we do not usually do when viewing a face. More recent systems, such as CD-fit, are far more flexible, allowing the position, size, and shape of individual features to be altered interactively on a computer. However, the essential problem of being a composite system remains. If you change, for example, the nose, our perception of the eyes and mouth may well change: we perceive faces holistically. The resultant images also look rather synthetic. The system described here produces face images by recombining principal components (eigenfaces), which are inherently global in nature and produce more photographic-like images.

Understanding the multiframe caricature advantage for recognizing facial composites

Eyewitnesses often construct a “composite” face of a person they saw commit a crime, a picture that police use to identify suspects. We described a technique (Frowd, Bruce, Ross, McIntyre, & Hancock, 2007) based on facial caricature to facilitate recognition of these images: Correct naming substantially improves when composites are seen with progressive positive caricature, where distinctive information is enhanced, and then with progressive negative caricature, the opposite. Over the course of four experiments, the underpinnings of this mechanism were explored. Positive-caricature levels were found to be largely responsible for improving naming of composites, with some benefit from negative-caricature levels. Also, different frame-presentation orders (forward, reverse, random, repeated) facilitated equivalent naming benefit relative to static composites. Overall, the data indicate that composites are usually constructed as negative caricatures.

Verbalization effects in facial composite production

Describing a persons face can temporarily interfere with face recognition ability. We explored whether this so-called ‘verbal overshadowing effect’ (VOE) might interfere with the construction of a traditional facial composite, a face produced by the selection of individual features: hair, eyes, mouth, etc. Participants looked at an unfamiliar target and two days later constructed a single composite after (a) describing the face (verbal no-delay), (b) without describing (no-description) or (c) 30 minutes after describing (verbal delay). Composite quality was overall of poor quality but it was worse in the verbal no-delay group relative to the no-description group, suggesting the involvement of a VOE, but equivalent between verbal delay and no-description, suggesting the presence of a ‘release’ from overshadowing. The data support the revised ‘transfer inappropriate retrieval’ explanation of the VOE, and suggest that witnesses to real crimes should not proceed directly from face description to feature selection.