Benton Test Of Facial Recognition

[Osman Briseno]

The Benton Facial Recognition Test (BFRT: Benton & Van Allen, 1968; see Benton et al. (1983) for the formal reference of the test) is a face matching task that is traditionally administered face-to-face using hard copy materials. Participants are simultaneously presented with a target face above an array of six test faces. In the first six trials, one face in the array matches the identity of the target face, and in the final 16 trials, three faces in the array match the identity of the target. The task was originally developed for the assessment of individuals believed to have acquired prosopagnosia (a severe deficit in recognising familiar people from their face) following brain injury (Barton, 2008; Bate & Bennetts, 2015; Van Belle et al., 2011), but has since been widely used to assess face perception skills in a number of neurological, clinical and psychiatric conditions (Annaz et al., 2009; Rabin et al., 2005; Sachse et al., 2014).

Another way to address this issue is to administer multiple versions of a face perception task, using rather different facial stimuli. This should prohibit, or at least reduce, the transfer of compensatory strategies that are useful in one version of a task. Indeed, some DPs report the use of particular facial features when images are captured within the same photography session (e.g. no change in skin tone or appearance of the hairline, even when images are cropped), or even consider pictorial cues such as lighting conditions or the quality of the images themselves (Adams et al., 2019). Further, very recent findings also highlight the importance of repeat testing on key measures of face recognition performance when screening for DP (Bate, Bennetts, Gregory, et al., 2019a; Murray & Bate, 2020), given issues with task reliability, the occurrence of borderline scores that are difficult to reconcile, and the possibility that a particular score simply occurred by chance performance (Young et al., 1993).

A new face matching task (the BFRT-r) was created that follows the original BFRT paradigm, but is computerised (akin to the BFRT-c) and uses new, more ambient facial images. We initially assessed the psychometric properties of the task and collected norming data from young typical adults. The reliability and validity of the BFRT-r was investigated by comparing performance on this task to the BFRT-c. In addition, content validity was assessed by comparing performance to (a) the Cambridge Face Memory Test (CFMT) and (b) to a new group of participants using an inverted version of the BFRT-r.

A total of 165 participants took part in Experiment 1. One hundred and nine participants aged between 18 and 35 years (mean age = 24.7 years, SD = 3.5; 55 female) completed the full string of tests in their upright format. To avoid re-exposure effects, 56 different participants aged between 18 and 35 years (mean age = 24.9 years, SD = 3.5; 27 male) completed only the inverted version of the BFRT-r. All participants were recruited via the online participant recruitment website Prolific, in exchange for a small financial incentive. All were Caucasian and lived within the UK, reported no history of socio-emotional, neurological or psychiatric disorder, and had normal or corrected-to-normal vision. This project was approved by the institutional Research Ethics Committee.

BFRT-c (Rossion & Michel, 2018): The BFRT-c is the original version of the BFRT, in a computerised format. The test contains a total of 22 trials in which an unfamiliar Caucasian target face (shown from a frontal viewpoint with a neutral expression) has to be found among a simultaneously presented array of six Caucasian probe faces, also showing neutral expressions. For the first six trials (half male), the target face has to be found only once within each array, where all faces are shown from a frontal viewpoint, such that the corresponding probe image is very similar to the target image. For the remaining 16 trials (half male), the target face is again presented from a frontal viewpoint. The participant is required to find three images within the six-image array that match the identity of the target. The six faces in each array vary either in terms of head orientation (the second section of the test: eight trials, half female) or lighting (the third section of the test: eight items, half female). Some target faces are repeated: four of the seven female targets appear in two separate sections, one of the seven male targets appears in all three sections, and three male targets are used in two sections. All target identities are also used as distractors in at least one trial of the task.

In each trial, target faces are presented at a slightly different size than those in the array (target faces were 156 x 232 pixels; faces in the array were 201 x 234 pixels, in order to minimise successful matching based on low-level, image-based visual cues: Rossion & Michel, 2018). All images are grayscale and display the overall shape of the face, but are cropped below the chin and beyond the hairline. As in the original version of the task, the order of the trials is not randomised and participants have an unlimited length of time to complete each trial. There is an inter-stimulus interval of 800 ms. Information screens at the beginning of each section instruct the participant how many responses to make for each trial, and inform them that response time is recorded.

Participants are required to select their responses by clicking on the appropriate face(s) in each array. For trials that require three responses, participants are able to select faces in any order, but cannot change a response once a face has been selected. The maximum score on the task is 54. Participants can receive one point in each of the six trials that compose Section 1 (where one response is required per trial), and between 0 and 3 points for each of the trials in Section 2 (where three responses are required per trial). Trial completion times are measured to aid data processing (see below), and overall task completion times are monitored for analysis.

A unique target was used in each of the 22 trials, and no target was re-used as a distractor. Ten distractor identities were repeated over the 22 trials, but different images of each individual were used where possible; only two images were repeated twice through the test. No distractor identity was repeated in the same array. Distractors were allocated to each trial based on their perceived similarity to the target, as judged by a member of the research team. Pilot testing supported these judgements: the trials included in the final BFRT-r did not elicit ceiling nor floor effects. In total, the test used images from 76 different individuals.

The CFMT is an unfamiliar face memory test. The overall objective of the CFMT is to introduce unfamiliar, young male faces to the participant and then test their recognition of those faces. It contains three test stages which increase in difficulty as the test progresses: (a) Learn: Participants view a target face from three viewpoints for 3 s per image. They then choose which of three faces, presented simultaneously, is the target. This is repeated for six faces, resulting in a maximum score of 18. (b) Test stage: Thirty triads of faces are presented, where one face is a novel image of a target identity intermixed with two distractors. (c) Noise stage: Twenty-four new triads are displayed with added visual noise. Again, each trial contains any one of the targets and two distractors. The entire test is scored out of 72, and chance is 24. More information about this test can be found in the associated publication. We include it here as a means to check the content validity of the BFRT-r (i.e. that it correlates well with a dominant face-processing task that is already known to have high content validity).

All tasks were completed online, using the Testable platform (www.testable.org; see Rezlescu et al., 2020). Participants were required to initially calibrate the tests for screen size, ensuring uniform presentation. The 109 participants that completed the main string of tests completed the CFMT, BFRT-r, and BFRT-c, in that order. This enabled us to collect accurate norming data for the new task without introducing practice effects from the repeated use of the same paradigm. The 56 participants who only took part in the inverted version of the BFRT-r did not complete any other tests.

Here, we present the BFRT-r: a new test of face perception that adopts the same paradigm as the original BFRT (as per the BFRT-c) but uses more naturalistic images to accommodate within-person variation in facial images. As the BFRT-r follows the procedure of the BFRT-c, the test continues to be simple and quick to administer, with an approximate completion time of four minutes in typical young adults. Initial analyses reveal that the BFRT-r has good internal reliability with strong inter-item correlations. It also has a strong inversion effect according to accuracy (although not completion times), suggesting that it taps face- rather than image-processing mechanisms. A strong correlation with the CFMT further supports this.

Comparison of performance on the two tasks indicates that the BFRT-r is slightly more difficult than the original version. More importantly, typical participants are able to score well above chance on the BFRT-r (the lowest score was 55.56%; chance performance is 46.30%). Further, the norming data reported here (M = 78.24%, SD = 9.20) would enable clinical participants to score two SDs below the mean without performing at chance level. Indeed, those with DP often show impaired face perception skills, but these skills are not completely abolished to the point that they are scoring at chance level (e.g. Bate et al., 2019a, b, c; Biotti et al., 2019; Righart & de Gelder, 2007). Thus, the task is suitably calibrated to detect variations in performance between chance and the control mean.