Face Matching Performance Research Articles

Expression-insensitive 3D face recognition by the fusion of multiple subject-specific curves

Abstract This study proposes a 3D face recognition method using multiple subject-specific curves insensitive to intra-subject distortions caused by expression variations. Considering that most sharp variances in facial convex regions are closely related to the bone structure, the convex crest curves are first extracted as the most vital subject-specific facial curves based on the principal curvature extrema in convex local surfaces. Then, the central profile curve and the horizontal contour curve passing through the nose tip are detected by using the precise localization of the nose tip and symmetry plane. Based on their discriminative power and robustness to expression changes, the three types of curves are fused with appropriate weights at the feature-level and used for matching 3D faces with the iterative closest point algorithm. The combination of multiple expression-insensitive curves is complementary and provides sufficient and stable facial surface features for face recognition. In addition, for each convex crest curve, an expression-irrelevant factor is assigned as the adaptive weight to improve the face matching performance. The results of experiments using two public 3D databases, GavabDB and BU-3DFE, demonstrate the effectiveness of the proposed method, and its recognition rates on both databases reflect an encouraging performance.

Comparing human and deep convolutional neural network face-matching performance on disguised face images

Comparing human and deep convolutional neural network face-matching performance on disguised face images

Face matching impairment in developmental prosopagnosia.

Developmental prosopagnosia (DP) is commonly referred to as 'face blindness', a term that implies a perceptual basis to the condition. However, DP presents as a deficit in face recognition and is diagnosed using memory-based tasks. Here, we test face identification ability in six people with DP, who are severely impaired on face memory tasks, using tasks that do not rely on memory. First, we compared DP to control participants on a standardized test of unfamiliar face matching using facial images taken on the same day and under standardized studio conditions (Glasgow Face Matching Test; GFMT). Scores for DP participants did not differ from normative accuracy scores on the GFMT. Second, we tested face matching performance on a test created using images that were sourced from the Internet and so varied substantially due to changes in viewing conditions and in a person's appearance (Local Heroes Test; LHT). DP participants showed significantly poorer matching accuracy on the LHT than control participants, for both unfamiliar and familiar face matching. Interestingly, this deficit is specific to 'match' trials, suggesting that people with DP may have particular difficulty in matching images of the same person that contain natural day-to-day variations in appearance. We discuss these results in the broader context of individual differences in face matching ability.

Causal evidence of the involvement of the right occipital face area in face-identity acquisition

There is growing evidence that the occipital face area (OFA), originally thought to be involved in the construction of a low-level representation of the physical features of a face, is also taking part in higher-level face processing. To test whether the OFA is causally involved in the learning of novel face identities, we have used transcranial magnetic stimulation (TMS) together with a sequential sorting – face matching paradigm (Andrews et al. 2015). First, participants sorted images of two unknown persons during the initial learning phase while either their right OFA or the Vertex was stimulated using TMS. In the subsequent test phase, we measured the participants’ face matching performance for novel images of the previously trained identities and for two novel identities. We found that face-matching performance accuracy was higher for the trained as compared to the novel identities in the vertex control group, suggesting that the sorting task led to incidental learning of the identities involved. However, no such difference was observed between trained and novel identities in the rOFA stimulation group. Our results support the hypothesis that the role of the rOFA is not limited to the processing of low-level physical features, but it has a significant causal role in face identity encoding and in the formation of identity-specific memory-traces.

An easy game for frauds? Effects of professional experience and time pressure on passport-matching performance.

Despite extensive research on unfamiliar face matching, little is known about factors that might affect matching performance in real-life scenarios. We conducted 2 experiments to investigate the effects of several such factors on unfamiliar face-matching performance in a passport-check scenario. In Experiment 1, we assessed the effect of professional experience on passport-matching performance. The matching performance of 96 German Federal Police officers working at Munich Airport was compared with that of 48 novices without specific face-matching experience. Police officers significantly outperformed novices, but nevertheless missed a high ratio of frauds. Moreover, the effects of manipulating specific facial features (with paraphernalia like glasses and jewelry, distinctive features like moles and scars, and hairstyle) and of variations in the physical distance between the faces being matched were investigated. Whereas manipulation of physical distance did not have a significant effect, manipulations of facial features impaired matching performance. In Experiment 2, passport-matching performance was assessed in relation to time constraints. Novices matched passports either without time constraints, or under a local time limit (which is typically used in laboratory studies), or under a global time limit (which usually occurs during real-life border controls). Time pressure (especially the global time limit) significantly impaired matching performance. (PsycINFO Database Record

Changing camera-to-subject distance changes face matching performance

Changing camera-to-subject distance changes face matching performance

Identity-level representations affect unfamiliar face matching performance in sequential but not simultaneous tasks.

According to cognitive and neurological models of the face-processing system, faces are represented at two levels of abstraction. First, image-based pictorial representations code a particular instance of a face and include information that is unrelated to identity-such as lighting, pose, and expression. Second, at a more abstract level, identity-specific representations combine information from various encounters with a single face. Here we tested whether identity-level representations mediate unfamiliar face matching performance. Across three experiments we manipulated identity attributions to pairs of target images and measured the effect on subsequent identification decisions. Participants were instructed that target images were either two photos of the same person (1ID condition) or photos of two different people (2ID condition). This manipulation consistently affected performance in sequential matching: 1ID instructions improved accuracy on "match" trials and caused participants to adopt a more liberal response bias than the 2ID condition. However, this manipulation did not affect performance in simultaneous matching. We conclude that identity-level representations, generated in working memory, influence the amount of variation tolerated between images, when making identity judgements in sequential face matching.

Developmental improvement and age-related decline in unfamiliar face matching.

Age-related changes have been documented widely in studies of face recognition and eyewitness identification. However, it is not clear whether these changes arise from general developmental differences in memory or occur specifically during the perceptual processing of faces. We report two experiments to track such perceptual changes using a 1-in- 10 (experiment 1) and 1-in-1 (experiment 2) matching task for unfamiliar faces. Both experiments showed improvements in face matching during childhood and adult-like accuracy levels by adolescence. In addition, face-matching performance declined in adults of the age of 65 years. These findings indicate that developmental improvements and aging-related differences in face processing arise from changes in the perceptual encoding of faces. A clear face inversion effect was also present in all age groups. This indicates that those age-related changes in face matching reflect a quantitative effect, whereby typical face processes are engaged but do not operate at the best-possible level. These data suggest that part of the problem of eyewitness identification in children and elderly persons might reflect impairments in the perceptual processing of unfamiliar faces.

Motion facilitates face perception across changes in viewpoint and expression in older adults.

Faces are inherently dynamic stimuli. However, face perception in younger adults appears to be mediated by the ability to extract structural cues from static images and a benefit of motion is inconsistent. In contrast, static face processing is poorer and more image-dependent in older adults. We therefore compared the role of facial motion in younger and older adults to assess whether motion can enhance perception when static cues are insufficient. In our studies, older and younger adults learned faces presented in motion or in a sequence of static images, containing rigid (viewpoint) or nonrigid (expression) changes. Immediately following learning, participants matched a static test image to the learned face which varied by viewpoint (Experiment 1) or expression (Experiment 2) and was either learned or novel. First, we found an age effect with better face matching performance in younger than in older adults. However, we observed face matching performance improved in the older adult group, across changes in viewpoint and expression, when faces were learned in motion relative to static presentation. There was no benefit for facial (nonrigid) motion when the task involved matching inverted faces (Experiment 3), suggesting that the ability to use dynamic face information for the purpose of recognition reflects motion encoding which is specific to upright faces. Our results suggest that ageing may offer a unique insight into how dynamic cues support face processing, which may not be readily observed in younger adults' performance. (PsycINFO Database Record (c) 2014 APA, all rights reserved).

Unfamiliar face matching: Pairs out-perform individuals and provide a route to training.

Matching unfamiliar faces is known to be difficult. Here, we ask whether performance can be improved by asking viewers to work in pairs, a manipulation known to increase accuracy for low-level visual discrimination tasks. Across four experiments we consistently find that face matching accuracy is higher for pairs of viewers than for individuals. This 'pairs advantage' is generally driven by adopting the response of the higher scoring partner. However, when the task becomes difficult, both partners' performance is improved by working in a pair. In two experiments, we find evidence that working in a pair can lead to subsequent improvements in individual performance, specifically for viewers whose accuracy is initially low. The pairs' technique therefore offers the opportunity for substantial improvements in face matching performance, along with an added training benefit.

Face drawing experience is associated with better face recognition performance and reduced left-side bias in face perception

Recent research in computer vision showed that while both analytic (local) and holistic (global) approaches perform well in face recognition, the fusion of the two yields the best performance. Professional face artists/caricaturists mastering both the drawing of individual facial features and whole faces may engage in such fused processing of faces, and thus may excel in face recognition. Zhou et al. (2012) found that art students with face drawing experience showed weaker holistic face processing than non-drawers; nevertheless, the two groups did not differ in sequential face-matching performance. Here we recruited professional face artists, in contrast to art students, in a simultaneous face-matching task with unfamiliar faces that differed in illumination, pose, disguise and expression. In contrast to sequential face-matching, our task reduces the involvement of visual short-term memory, as face artists' advantage is more likely to be in their perceptual ability than in memory. We recruited 15 face artists and 15 non-drawers. Before the experiment, participants drew a face within 15 minutes. An experienced face artist rated participants' expertise according to the drawings. We found that face artists outperformed non-drawers in the simultaneous face-matching task (t(28) = 2.56, p <.05). Face drawing expertise was positively correlated with face recognition performance (r = 0.38, p <.05). In addition, while non-drawers judged a face made from two left half-faces more similar to the original face than one from two right half-faces, i.e., the left side bias (right hemisphere dominance) in face perception, face artists had a reduced bias than non-drawers (t(28) = -2.39, p <.05), suggesting reduced global face processing. Face drawing expertise was negatively correlated with the bias (r = -0.47, p <.05). Together, the results suggest that face drawing expertise may shift attention from global to local, which may help face recognition in challenging tasks. Meeting abstract presented at VSS 2014

Redesigning photo-ID to improve unfamiliar face matching performance.

Viewers find it difficult to match photos of unfamiliar faces for identity. Despite this, the use of photographic ID is widespread. In this study we ask whether it is possible to improve face matching performance by replacing single photographs on ID documents with multiple photos or an average image of the bearer. In 3 experiments we compare photo-to-photo matching with photo-to-average matching (where the average is formed from multiple photos of the same person) and photo-to-array matching (where the array comprises separate photos of the same person). We consistently find an accuracy advantage for average images and photo arrays over single photos, and show that this improvement is driven by performance in match trials. In the final experiment, we find a benefit of 4-image arrays relative to average images for unfamiliar faces, but not for familiar faces. We propose that conventional photo-ID format can be improved, and discuss this finding in the context of face recognition more generally.

Integration of internal and external facial features in 8- to 10-year-old children and adults

Investigation of whole-part and composite effects in 4- to 6-year-old children gave rise to claims that face perception is fully mature within the first decade of life (Crookes & McKone, 2009). However, only internal features were tested, and the role of external features was not addressed, although external features are highly relevant for holistic face perception (Sinha & Poggio, 1996; Axelrod & Yovel, 2010, 2011). In this study, 8- to 10-year-old children and adults performed a same–different matching task with faces and watches. In this task participants attended to either internal or external features. Holistic face perception was tested using a congruency paradigm, in which face and non-face stimuli either agreed or disagreed in both features (congruent contexts) or just in the attended ones (incongruent contexts). In both age groups, pronounced context congruency and inversion effects were found for faces, but not for watches. These findings indicate holistic feature integration for faces. While inversion effects were highly similar in both age groups, context congruency effects were stronger for children. Moreover, children's face matching performance was generally better when attending to external compared to internal features. Adults tended to perform better when attending to internal features. Our results indicate that both adults and 8- to 10-year-old children integrate external and internal facial features into holistic face representations. However, in children's face representations external features are much more relevant. These findings suggest that face perception is holistic but still not adult-like at the end of the first decade of life.

Infrequent identity mismatches are frequently undetected

The ability to quickly and accurately match faces to photographs bears critically on many domains, from controlling purchase of age-restricted goods to law enforcement and airport security. Despite its pervasiveness and importance, research has shown that face matching is surprisingly error prone. The majority of face-matching research is conducted under idealized conditions (e.g., using photographs of individuals taken on the same day) and with equal proportions of match and mismatch trials, a rate that is likely not observed in everyday face matching. In four experiments, we presented observers with photographs of faces taken an average of 1.5 years apart and tested whether face-matching performance is affected by the prevalence of identity mismatches, comparing conditions of low (10 %) and high (50 %) mismatch prevalence. Like the low-prevalence effect in visual search, we observed inflated miss rates under low-prevalence conditions. This effect persisted when participants were allowed to correct their initial responses (Experiment 2), when they had to verify every decision with a certainty judgment (Experiment 3) and when they were permitted "second looks" at face pairs (Experiment 4). These results suggest that, under realistic viewing conditions, the low-prevalence effect in face matching is a large, persistent source of errors.

Face Tracking and Recognition at a Distance: A Coaxial and Concentric PTZ Camera System

Face recognition has been regarded as an effective method for subject identification at a distance because of its covert and remote sensing capability. However, face images have a low resolution when they are captured at a distance (say, larger than 5 meters) thereby degrading the face matching performance. To address this problem, we propose an imaging system consisting of static and pan-tilt-zoom (PTZ) cameras to acquire high resolution face images up to a distance of 12 meters. We propose a novel coaxial-concentric camera configuration between the static and PTZ cameras to achieve the distance invariance property using a simple calibration scheme. We also use a linear prediction model and camera motion control to mitigate delays in image processing and mechanical camera motion. Our imaging system was used to track 50 different subjects and their faces at distances ranging from 6 to 12 meters. The matching scenario consisted of these 50 subjects as probe and additional 10 000 subjects as gallery. Rank-1 identification accuracy of 91.5% was achieved compared to 0% rank-1 accuracy of the conventional camera system using a state-of-the-art matcher. The proposed camera system can operate at a larger distance (up to 50 meters) by replacing the static camera with a PTZ camera to detect a subject at a larger distance and control the second PTZ camera to capture the high-resolution face image.

Acute Anxiety Impairs Accuracy in Identifying Photographed Faces

We investigated whether acutely induced anxiety modifies the ability to match photographed faces. Establishing the extent to which anxiety affects face-matching accuracy is important because of the relevance of face-matching performance to critical security-related applications. Participants (N = 28) completed the Glasgow Face Matching Test twice, once during a 20-min inhalation of medical air and once during a similar inhalation of air enriched with 7.5% CO2, which is a validated method for inducing acute anxiety. Anxiety degraded performance, but only with respect to hits, not false alarms. This finding provides further support for the dissociation between the ability to accurately identify a genuine match between faces and the ability to identify the lack of a match. Problems with the accuracy of facial identification are not resolved even when viewers are presented with a good photographic image of a face, and identification inaccuracy may be heightened when viewers are experiencing acute anxiety.

Face Processing Changes in Normal Aging Revealed by fMRI Adaptation

We investigated the neural correlates of facial processing changes in healthy aging using fMRI and an adaptation paradigm. In the scanner, participants were successively presented with faces that varied in identity, viewpoint, both, or neither and performed a head size detection task independent of identity or viewpoint. In right fusiform face area (FFA), older adults failed to show adaptation to the same face repeatedly presented in the same view, which elicited the most adaptation in young adults. We also performed a multivariate analysis to examine correlations between whole-brain activation patterns and behavioral performance in a face-matching task tested outside the scanner. Despite poor neural adaptation in right FFA, high-performing older adults engaged the same face-processing network as high-performing young adults across conditions, except the one presenting a same facial identity across different viewpoints. Low-performing older adults used this network to a lesser extent. Additionally, high-performing older adults uniquely recruited a set of areas related to better performance across all conditions, indicating age-specific involvement of this added network. This network did not include the core ventral face-processing areas but involved the left inferior occipital gyrus, frontal, and parietal regions. Although our adaptation results show that the neuronal representations of the core face-preferring areas become less selective with age, our multivariate analysis indicates that older adults utilize a distinct network of regions associated with better face matching performance, suggesting that engaging this network may compensate for deficiencies in ventral face processing regions.

Developmental changes in the microgenesis of face perception revealed by effects of context and inversion

Present studies on the development of face perception mechanisms are ambiguous about the question of whether holistic face vision arises early, or in the second decade of life ( Crookes & McKone, 2009). Measuring the time course of face matching we assess effects of context and inversion as correlates of holistic processing in the microgenesis of face perception within the first 650 ms, and compare among 8- to 10-year-old children and adults. Results for adults indicate dominance of holistic viewing at brief timings, which is gradually replaced by feature selective strategies enabling them to selectively attend either internal or external features, as demanded by instruction. For children, however, effects of context and inversion are absent at brief timings, but gradually increase to strong levels with increasing viewing times. Moreover, we find a pronounced asymmetry in face matching performance with internal and external features. While face matching by attending external features is well developed and robust against variable facial contexts, face matching by attending internal features is generally poor, and strongly affected by interleaved congruent and incongruent contextual information. These results indicate that children and adults differ not only in the kind of featural information they preferentially encode in face perception, but also in the processing time they need to build holistic representations. While these are fast and automatic in adults’ face vision, children’s face representations are part based at brief timings, but develop to integrated wholes as more temporal resources are made available.

The time course of face matching for featural and relational image manipulations

It was found recently that horizontal and vertical relationships of facial features are differently vulnerable to inversion (Goffaux & Rossion, 2007). When faces are upside down manipulations of vertical relations are difficult to detect, while only moderate performance deficits are found for manipulations of horizontal relations, or when features differ. We replicate the findings of Goffaux and Rossion, and record the temporal courses of face matching performance and the effects of inversion. For vertical relations and featural changes inversion effects arise immediately, starting with the first 50 ms of processing. For horizontal relations inversion effects are absent at brief timings, but arise later, after about 200 ms. The final magnitudes of inversion effects are different in all three conditions, reaching 23.4% for vertical displacement, 10.7% for feature replacement, and 5.8% for horizontal displacement. Our results underline the special status of horizontal in contrast to vertical feature relations, and indicate that local and global configural information is handled in parallel by specific routines, as proposed recently (Senunkova & Barton, 2008).

The effect of homeomorphic image transformations on face matching performance

The effect of homeomorphic image transformations on face matching performance