Unfamiliar Face Matching Task and Multiple Image Benefit

Word count: 2500

Student: Shashnavee Deepa D/o Sangkuni

Student Number:

Subject: PSYC327

Tutor: Dr Li Mei Chew

Class: Monday, 3.30 – 6.30pm

Human unfamiliar face matching is prone to errors, but some research indicates that using

numerous image arrays rather than just one could lead to better results. In this study, we

investigate whether matching to target images of a face rotated in view benefits from within-

person heterogeneity from front views. Here, one or three-picture arrays with high or low

variability and a target image selected from the high and low variability image sets were shown.

Three investigations used either AFC or simultaneous presentation of the arrays, as well as AFC

or simultaneous presentation of the target image. The only AFC viewing that improved overall

accuracy across trial types, especially for high variability arrays, was watching a multiple-image

array followed by a target picture with high variability. When the target image and array objects

were similar, accuracy was at its highest. The findings highlight the significance of picture

similarity and indicate that variability is most useful when the array and target are presented in

order.
 Unfamiliar Face Matching Task and Multiple Image Benefit

Pictures are frequently used to establish a person's identity or to confirm that they are whom they

say they are. The function of background pictures in the process of familiarity has been the

subject of face recognition literature. A method of recognising or verifying a person's

identification using their face is facial recognition. This is because people are accustomed to

seeing familiar faces in a variety of poses, expressions, and lighting, which makes it possible to

extract stable features and make recognition reliant on representations that summarise within-

person variability (differences between various perspectives on the same face) (White et al.,

2014). Matching unfamiliar faces is difficult as it is found to have superior unfamiliar face

matching abilities, as the face matching ability may be precise. Constructing a more abstract

picture of the face and preserving knowledge about within-person variability has proven to

improve novel matching tasks when various photo arrays and ambient images are presented

(Hunnisett & Favelle, 2021). This highlights the need to look into techniques to enhance

unknown face recognition, a concept commonly known as the multiple image advantage.

Personally familiar faces are processed more intensely and effectively compared to unfamiliar

faces. The human face processing system comprises a core system visually examining faces and

a comprehensive system for retrieving person-knowledge and other nonvisual data. (Chapman et

al., 2018) Familiar faces are remembered so well because of how many different situations and

image modifications they have been exposed to, such as array size or amount of image

variability. Unfamiliar faces do not benefit from this depth of experience and are represented

using information unique to the individual image, are harder to recognise, and take longer to

identify.
In the simultaneous matching task, all the information necessary to choose the response is on

each trial presented simultaneously so that no form of retention is required (Gelder & Bertelson,

2009). A sequential matching task can simultaneously capture essential information in a context

and model relationships among utterances (Wu et al., 2019). Ritchie et al. (2021) looked into the

impact of four-image arrays in simultaneous and sequential matching tasks. They discovered that

the sequential matching task had an advantage over the simultaneous matching problem due to

numerous images. Participants must use memory to preserve the representation created from the

multiple image array shown first and then compare it against a target in sequential matching

tasks to abstract a stable representation from the underlying unpredictability in the various

images. Face associations with unfamiliar identities displayed with high or low within-person

variability, i.e., the same person's image appearing when a search for their name is done online

(high variability) or multiple photographs of the same person taken from the same event (low

variability) (Ritchie & Burton, 2017).

In a related study, Sandford and Ritchie (2021) compared a target image to an array with high or

low variability in simultaneous and sequential matching tasks. The only sequential viewing that

improved overall accuracy across trial types was watching a multiple-image array followed by a

target picture with high variability. In a sequential matching test, White et al. (2014) prompted

participants to match another image to an array of two, three, or four photographs of the same

person. The multiple-image arrays beat single-image matching, demonstrating a multiple-image

images with a target image in front, three-quarter or profile view was examined in a study by

Hunnisett and Favelle (2021), which required participation in either a simultaneous or sequential

matching task. Multiple picture benefits were detected in both tasks for all three angles of view,

while a larger significant effect was found in the sequential matching task.

Another explanation would be the tactics used could be another explanation for the existence of

several image advantages in concurrent activities. Participants use the closest match technique

when the picture array and target are shown side by side in pairwise comparison, and as the

likelihood of finding an image similar to the target image increases (particularly as image array

size increases), so does the precision of matching performance (Menon et al., 2015; Ritchie et al.,

2021). Thus, the closest match method could be responsible for several image advantages in

concurrent jobs.

The two-alternative forced choice (2AFC) method uses the observer's pattern of choices and

response times to two copies of the sensory input to gauge how sensitive the person is to that

particular sensory input or stimulus (Ritchie et al., 2021). Further studies like (Ratcliff et al.,

2018) examined participant performance on five perceptual 2-alternative forced-choice (2AFC)

tasks, in which two stimuli are presented, and a comparison judgment is to be made. The

magnitudes of the two stimuli and their difference can affect how the information represented in

these 2AFC tasks is chosen. Participants replied more quickly when the combined magnitude of

the increased array size. These results support the idea that the overall image array size

influences decision-making.
This research uses a 2x2x2 factorial design with task type, array type and array size as the

variables. This study hypothesised that there would be no difference in performance between the

1 or 3 picture arrays and the high or low image variabilities for the simultaneous same or

different matching task. Lastly, it was hypothesised that in the simultaneous 2AFC matching

task, the three image arrays and high variability images would produce a better performance.

Method

Participants.

Participants were 50 undergraduate students from the University of Wollongong. (37 females,

mean age 22.4 yr, SD = 3.28)

Stimuli and apparatus.

The stimuli used in the experiment were face images sourced from a database created by

researchers at the University of New South Wales. Images were 80 caucasian identities. Images

were sourced from a database created by researchers at the University of New South Wales. The

stimuli selected for use in this study for each identity were one target image, 7 HV images OR 7

LV images (8 images total for each identity), for 640 images. Target images were captured in

controlled conditions. High variability images were ambient images sourced from Facebook. The
ambient images varied in pose, expression and lighting. Low variability images were frames

taken from a video of a person talking. The images had similar lighting, clothing etc. All images

were cropped to 250 x 300 pixels. Participants performed the experiment on a Dell Optiplex

7440 computer with a 48-cm flat-screen Dell monitor with a screen resolution of 1920! 1080

pixels. Superlab Pro 5.0 (Cedrus corporation, 2013) was used to control the presentation and

record responses and reaction times.

Design

A two-way mixed ANOVA with the within-subjects factors was conducted. The independent

variables were, Trial Type (same, different, 2AFC), image variability (High and Low) and 2

Array Size (1 and 3). The dependent variable was the accuracy of face recognition performance.

A factorial design was used for the experiment trials. The alpha value used was .05.

Procedure

Each participant received the trials in a different random order adjustment of all factors within

subjects. First, for match trials, a 2 (variability: high, low) x 2 (number of images: 1, 3) , the

matching challenge's objective could be either a match to the array or a foil (mismatch). The foil

identity shared the target identity's gender, eye colour, and hair colour. For the trial events,

central fixation cross (250 ms); Stimuli delivered with the target stimulus on the left and the

array (one or three image/s) shown on the right in a triangle configuration; Stimuli exhibited

until response was produced.

The target and foil were the two options in the 2AFC task where the conditions remained except

that the display showed the target and foil pair of stimuli on the left of the screen with the array

presented on the right. The Inter trial interval was 500 milliseconds for both tasks. The task was
blocked and the order of the blocks was counterbalanced across participants, and the display

remained on the screen until a response was made. There were six trials for the same/different

task, and 4 for 2AFC. The target and foil pair of stimuli were presented on the left side of the

screen, while the array was presented on the right, in the same trial events for the 2AFC task.

Until an answer was given, the display stayed visible on the screen.

Results

Normality assumption was violated p >.05. As such, a parametric test of Analysis of

Variance (ANOVA) was conducted as it is robust to violations of normality.

Figure 1

Bar Graph of the Mean proportion of correct data for the same/different trials are plotted in the

bar graph.

Note: Error bars represent ±1 standard error of the mean.

A repeated measures ANOVA was conducted to test accuracy in simultaneous same/different

face-matching tasks. The ANOVA indicated no effect on array size, F(1, 49) = 0.565, p=.456,

η2=.011, but a significant effect on image variability F(1, 49) = 5.338, p=.025, η2=.098. This

indicated that there wasn't any difference noticed between one image array (M = 2.40 , SD

=0.840) and three image arrays (M = 2.46 , SD = 0.873). This also indicates that there was a

difference noticed between high variability (M = 2.52 , SD = 0.78) and low variability (M = 2.35

, SD = 0.92). The interaction between array size and image variability was not significant, F(1,

49) = .306, p=.583, η2=.006

Figure 2
Mean d′ values as a function of target view and array size for the simultaneous face-matching

task in Experiment 1

Note: Error bars represent ±1 standard error of the mean.

Experiment 2

A repeated measures ANOVA was conducted to test accuracy in simultaneous

same/different face-matching tasks. The ANOVA indicated no effect on array size, F(1, 49) =

51.07, p=.000, η2=.510, but a significant effect on image variability F(1, 49) = 26.06, p=.000,

η2=.347. This indicated that there wasn't any difference noticed between one image array (M =

2.79 , SD = 1.21) and three image arrays (M = 3.65 , SD = 1.23). This also indicates that there
was a difference noticed between high variability (M = 3.54 , SD = 1.25) and low variability (M

= 2.91 , SD = 1.26). The interaction between array size and image variability was significant,

F(1, 49) = 14.038, p=.000, η2=.223

A post-hoc paired t-test was conducted. The difference in accuracy for high and low

variability for one array size was insignificant, t(49) = 1.53, p=.134. Whereas the difference in

accuracy for high variability and low variability for 3 array size was significant, t(49) = 6.682,

p=.000. The difference between high variability for 3 array size and high variability for 1 array

size was significant, t(49) = 8.438, p = .000. Lastly, The difference between low variability for 3

array size and low variability for 1 array size was significant, t(49) = 2.923, p = .005.

General Discussion

Matching unfamiliar faces is challenging since the face matching skill may be specialised,

leading to superior unfamiliar face matching abilities. Importantly, these results demonstrated

that the sequential matching paradigm is appropriate for testing the same/different matching task

effect, 2AFC, and extended the effect in the sequential presentation. This paradigm creates an

opportunity to systematically compare unique presentation sequences (sequential or

simultaneous matching) to elucidate the same/different matching task effect and 2AFC face

perception. Although we should note that our "low variability" image set includes a more

excellent range of variability than is often encountered in face learning experiments, we have

compared learning from high variability to learning from low variability in all the research

provided here.
In many previous studies investigating face matching, the experimental modification changed the

same/different performance, but not often in most earlier investigations on face matching (White

et al., 2014; Menon et al., 2015). White et al. (2014) observed advantages solely for match trials

in experiments comparing the advantages of average images versus arrays of images for face

matching tasks. This supports our first prediction patriotically. Before testing with a fresh batch

of images, the training technique ensures that participants have correctly learnt each of the initial

images of each identity; thus, this technique was avoided. Keeping the number of exposures

constant across identities and situations is essential to our approach, but this is not always

possible when using training techniques. Additionally, we did not want participants to memorise

specific representations of the identities; instead, we encouraged them to create mental images of

each unfamiliar individual to learn them. We contend that this method more closely resembles

real-world face learning.

Our current study focused on whether exposure to multiple (high variability) images in a

comparison array enhances performance on matching tasks. Our research demonstrate that for

targets displayed at the front, three-quarter, and profile viewpoints, multiple image arrays can

outperform a single image in terms of unfamiliar face-matching performance. Despite little

interaction, planned comparisons revealed that this improvement was considerable for front and

three-quarter view targets but not for profile-view targets. Multiple image arrays were the same

for unfamiliar face-matching performance regarding a one-image array or a three-image array.

There are a few methodological limitations in this study. The first limitation is the well-

documented "other-race effect," in which viewers find it easier to recognize faces of their race

than faces of other races, which is a very prevalent issue for face matching. Another limitation is
that we are so adept at identifying known faces in daily life; one explanation for why we have

faith in eyewitness identification is that we tend to over-generalize our confidence in these skills

by presuming that we are equally adept at identifying new faces. This is incorrect as

understanding the complexity of routine face recognition has significantly advanced thanks to

research into the psychology of face perception. It has also exposed fundamental flaws in

people's and machines' capacities for facial recognition.

Our findings have implications for enhancing face recognition accuracy in less-controlled

CCTV photos and for enhancing unknown face identification in a real-world setting. For the face

learning process to be successful and to create a solid identity-level face representation,

gathering and synthesizing information regarding the idiosyncratic variability of a face appears

to be essential. Future research can solve many problems with the theory and practice of

unknown face recognition by learning more about the contributions that within-person variation

makes to the representation of a face and what each encounter makes to the matching procedure.

Participants in our learning process acquire the ability to accept and cope with large levels of

intraindividual variation without this ultimately leading to a general acceptance of all images,

including images of foil identities. This reflects the expanding interest in and extent of the

research of within-person variability. In conclusion, in psychology tests, variations in a person's

face across images have long been considered a bothersome variable whose impact should be

reduced before data are collected.

Burton, A. M., Jenkins, R., Hancock, P. J. B., White, D. (2005). Robust representations

for face recognition: The power of averages. Cognitive Psychology, 51(3), 256–

284. https://doi.org/10.1016/j.cogpsych.2005.06.003

Chapman, A. F., Hawkins-Elder, H., & Susilo, T. (2018). How robust is familiar face

recognition? A repeat detection study of more than 1000 faces. Royal Society

Open Science, 5(5), 170634. https://doi.org/10.1098/rsos.170634

Gelder, B., & Bertelson, P. (2009). A Comparitive Approach To Testing Face Perception:

Face and Object Identification By Adults In A Simultaneous Matching Task (pp.

177–190).

Hunnisett, N., & Favelle, S. (2021). Within-person variability can improve the

identification of unfamiliar faces across changes in viewpoint. Quarterly Journal

of Experimental Psychology, 74(11), 1873–1887.

https://doi.org/10.1177/17470218211009771

Kaspersky. (2022, February 9). What is Facial Recognition – Definition and Explanation.

www.kaspersky.com. Retrieved September 15, 2022, from

https://www.kaspersky.com/resource-center/definitions/what-is-facial-recognition

https://doi.org/10.1080/17470218.2014.990468
Ratcliff, R., Voskuilen, C., & Teodorescu, A. (2018). Modeling 2-alternative forced-

choice tasks: Accounting for both magnitude and difference effects. Cognitive

Psychology, 103, 1–22. https://doi.org/10.1016/j.cogpsych.2018.02.002

Menon, N., White, D., & Kemp, R. I. (2015). Identity-level representations affect

unfamiliar face matching performance in sequential but not simultaneous tasks.

Quarterly Journal of Experimental Psychology, 68(9), 1777–1793.

https://doi.org/10.1080/17470218.2014.990468

Ritchie, K. L., & Burton, A. M. (2017). Learning faces from variability. Quarterly

Journal of Experimental Psychology, 70(5), 897–905.

https://doi.org/10.1080/17470218.2015.1136656

Ritchie, K. L., Kramer, R. S., Mileva, M., Sandford, A., & Burton, A. M. (2021).

Multiple-image arrays in face matching tasks with and without memory.

Cognition, 211, 104632. https://doi.org/10.1016/j.cognition.2021.104632

White, D., Burton, A. M., Jenkins, R., & Kemp, R. I. (2014). Redesigning photo-ID to

improve unfamiliar face matching performance. Journal of Experimental

Psychology: Applied, 20(2), 166–173. https://doi.org/10.1037/xap0000009

Wu, Y., Wu, W., Xing, C., Xu, C., Li, Z., & Zhou, M. (2019, March). A Sequential

Matching Framework for Multi-Turn Response Selection in Retrieval-Based

https://doi.org/10.1162/coli_a_00345
 Reflection

The assignment required us to look at each lab report into sections and analyze how they

logically format their information so the reader can comprehend and use their results. Next, I

outlined each of my chosen articles, making thorough notes on what was and was not included in

each report according to the feedback given by the tutor. With all this information outlined, I

wrote my first intro draft by going through each section of what should be included in a lab

report. Trying to condense all this data into something comprehendible was difficult, finding

similarities and differences was not the only difficult task. Making even one comparison or

showing one difference could take many sentences because some analysis of data was needed

beyond what was offered in the reports. I can now quickly analyze a lab report for the needed

information and judge if a report is usable for my research by glancing over the cover. In

conclusion, this analysis was excellent practice for writing and analysis, but also for researching.

I can now break down an article much faster because I know what to look for and where to find

it.