An Investigation of the Memory Coding of Facial Features: Are facial Features Coded in Accordance with the Distinctiveness Hypothesis.

by Ken Campbell
Missouri Western State College
May 1, 1997

Abstract

This study was designed to examine the distinctiveness hypothesis, as it relates to memory for faces, which states that distinctive features are stored in exaggerated form. Based on this concept the caricature hypothesis predicts that caricatures are better representations of images stored in memory. A group of 23 intermediate psychology students participated in a recall task designed to test these two concepts. The images used in this study were developed using a computerized image generator. The participants recall for faces was tested by first presenting a series of 8 pictures, 4 caricatures and 4 facial images. Four of the 8 were female and the other 4 were male. One week later, participants were shown a second series of 12 pictures, 6 female and 6 male. Some of the pictures were presented exactly as they appeared in the first series, and some that were initially presented as caricatures were presented as normal facial images. Some of the pictures that had appeared first as facial images were later presented as caricatures. There were also caricatures and facial images shown that had not appeared in any context in the first series. Participants were asked to indicate if the person depicted in the image had appeared in the original eight pictures. A 3X2X2 repeated measures ANOVA was conducted to analyze the data. There was a main effect found for what participants saw first, F (2, 44) = 6.94, p&LT .01. There was a two-way interaction found for what was presented first and what was shown second,F (2, 44) = 15.29,p .01. Participants were better able to identify images that were presented in the same form across all conditions. These results do not support the distinctiveness nor the caricature hypotheses.

Introduction

Most of the research studies on facial features and memory coding are based on two assumptions. First, the distinctiveness hypothesis which states that facial features are coded in memory with distinctive features exaggerated. Second , the caricature hypothesis which implies that facial caricatures serve as a better representation of the stored memory (Tversky, Baratz, 1984, Rhodes, Brennan, Carey, 1987). The study developed by Tversky and Baratz used faces of current political figures. Each face was presented as a spontaneous photograph and in caricature form. All the caricatures were developed by an artist who specialized in the development of caricatures. Participants were presented with a series of photographs and caricatures each with a name presented by it. In a timed task the participants were asked to indicate whether the name and the face matched. The results indicate that there is no correlation between the type of picture presented and the recall of names. The researchers noted that photographs produced the highest number of correct responses.

In review of the study by Tversky and Baratz (1984), researchers Rhodes, Brennan, and Carey (1987) suggest that photographs may show an advantage over caricatures because they provide more spatial information. To compensate for the inconsistencies of hand drawn caricatures these researchers employed a computerized caricature generator, developed by Brennan, to create a series of caricatures and anticaricatures from a veridical digitized line drawings. Seven drawings were drawings were created for each face with levels of exaggeration in proportions were .25, .50, .75 for caricatures and de-exaggerated for anticaricatures in proportions of -.25, -.50, -.75. The researchers conducted an identification task using participant who were familiar with the individuals depicted in the pictures. The results of the identification task did not support the assumption that the caricatures should be as recognizable as the veridical drawings. There was some support for the caricature and distinctiveness hypothesis as caricatures were identified quicker then the veridical drawings.

In a study conducted by Rhodes and Wooding (1989) the researchers developed a procedure to examine the speed and accuracy of photograph and caricature matching in both the right and left hemispheres of the brain. The stimuli used in this study consisted of caricatures and photographs of seventeen famous faces. The stimuli were varied in presentation in the right and left visual fields respectively. After hearing the name of a famous person participants were shown either a photograph or a caricature for 500 msec and asked to in determine if the name matched the face. The results of this study indicate that the right hemisphere is faster and more accurate in determining names and faces that did not match, however, there were no differences in speed and accuracy when names and faces matched.

Rhodes and Moody offer two possible explan ations for a caricature advantage shown for familiar faces. First, it is possible that distinctive information is exaggerated in long-term memory similar to a caricature. Second, caricatures are recognized quicker because the lack of details provides for less distraction and an increased emphasis on distinct features. In this study the researchers were looking to see if a caricature advantage existed using unfamiliar faces. A portion of the stimuli developed by Rhodes, Brennan, and Carey (1987) were used with participants who were unfamiliar with the faces. The procedure consisted of a learning task where they were shown photographs of faces. Later the participants were shown a series of drawings with the distinctive features varied in proportions ranging from -.50%, -.25%, 0, .25%, .50% and responses were based on whether the individual depicted in the drawing had been presented in the learning task. The results provided no evidence to support a caricature advantage with unfamiliar faces. The researchers indicate that these results support the idea that distinctive features are stored in long-term memory similar to caricatures.

If the caricature hypothesis is true, then a caricature in which only distinctive features are exaggerated should be easier to recognize then the original face. Based on this assumption Mauro and Kubovy (1992) designed a procedure which included a learning task in which slides containing faces and caricatures were presented. In a test procedure participants were presented with faces and caricatures previously seen, new faces and caricatures, and faces of caricatures previously seen and caricatures of faces previously seen. Results of this study show that caricatures were better reminders of the previously seen faces and, false caricatures were identified more reliably than false faces.

A study conducted by Benson and Perrett (1991) further developed the process of creating computer generated caricatures, first developed by Brennan (1984), by adding detail to caricatures using pixel-based images. In a rating task, participants were asked to determine which degree of exaggeration served as the best likeness of the target face. An assessment of the rating task indicates that positive caricatures of familiar faces serve as an enhanced likeness of the target faces. The results using unfamiliar faces were not as positive.

In a study performed by Stevenage (1995), a procedure was developed to test children in regard to the caricature hypothesis. The participants were divided into two groups, one group where presented with veridical faces and the other with caricature faces. Each face was matched with a fictitious name and a learning task was performed where children were required to learn a name for each face. Results of the study indicate that it took fewer trials to learn the names when presented with a caricatured face.

The purpose of this study is to re-examine the nature of the distinctiveness and caricature hypotheses using unfamiliar faces and caricatures with a memory identification task. The distinctiveness hypothesis suggests that facial images presented first, then presented second as caricatures should be identified more accurately. The caricature hypothesis suggests that caricatures are better representations of the stored memory of a face. The research hypothesis developed for this study suggests that caricatures presented second will provide higher recall accuracy for facial images presented first.

Methods

Subjects

Twenty-three intermediate level psychology students at Missouri Western State College participated in this study to receive extra credit for the course. Participants were assigned to identical tasks for a within-subject analysis.

Materials

The stimulus presented consisted of computer generated faces and caricatures with a spatial resolution of 360 X 360 pixels. The images were produced using Kai's Power Goo Realtime Liquid Image Funware . The caricatures were developed by exaggerating the most distinctive feature(s) as determined by students in a experimental psychology laboratory class. The images that were presented included, six male and six female faces. Caricatures were developed for three of the male faces and three of the female faces. Six conditions were designed from the twelve images for each sex. For condition one, a normal image of a face was developed. For condition two, a caricature of a separate face was developed. Condition three required that a normal image and a caricature be developed for a face. Condition four also required for the development of a normal image and caricature but for a different face than the one used in condition three. Condition five required the development of a facial image and condition six involved the production of another caricature.

Procedure

This study consisted of both a learning procedure and a recognition task. In the learning procedure participants were presented with eight images. These eight images included two male and two female normal facial images and, two male and two female caricatures. Participants were told that they would later be asked to identify the person depicted in each image. The images were presented one at a time on an over-head projector with a viewing time of 20 seconds for each image. One week later the participants were shown a total of 12 images for a period of 15 seconds each. As each image was presented, the participants were asked to indicate, by checking "yes" or "no," if the person depicted in the image had appeared in the learning task. The twelve images presented were representations of twelve different faces and provided for six different conditions. The image presented in condition one was a facial image presented in the learning task. The image presented in condition two was a repeat of one of the caricatures presented in the learning task. Condition three involved presenting a caricature that represented one of the facial images shown in the learning task. In condition four a facial image was presented that represented one of the caricatures shown in the learning task. Condition five involved presenting a facial image that was not shown in the learning task. Condition six involved presenting a caricature that had not been shown in the learning task. Each of the six conditions were presented as both male and female images.

Results

To look for main effects for what was seen first, what was seen second, the sex of the images, and to look for the presence of an interaction(s) amongst the three variables a 3X2X2 repeated measures ANOVA was conducted. There was a significant difference between the mean scores for correctly identifying images that had not appeared first, facial images that had appeared first, and caricatures that had appeared first, F (2, 44) = 6.94,p&LT.01.

A main effect was not found for what the participants saw second, a normal facial image or a caricature, F (1, 22) = .11,p&GT.05. No main effect was found when analyzing the mean scores in respect to the sex of the images, F (1, 22) = .49, p&GT.05.

There was an interaction between what the participants saw first and what was shown second,F (2, 44) = 15.29,p&LT.01. When participants were shown a facial image first, and the same facial image second they made correct decisions 96% of the time. When participants were shown a caricature first, and the same caricature second correct decisions were made 89% of the time. Caricatures that had not appeared first but were presented second were correctly identified as not seen before by 89% of the participants. Facial images that had not appeared first but were presented second were correctly identified as not seen before by 89% of the participants. When participants were shown a caricature first, and then second, the facial image used to develop the caricature participants responded correctly only 61% of the time. When participants were shown a facial image first then a caricature of that image second they made correct responses 70% of the time (figure 1).

There was no interaction between what the participants saw first and the sex of the images,F (2, 44) = .75,p&GT.05. Upon analyzing the data for what participants saw second and the sex of the image no interaction was found, F (1, 22) = .56, p&GT.05.

When the three factors, what they saw first, saw second, and the sex of the image, were analyzed no interaction was found, F ( 2, 44) = .83, p&GT.05. The analysis of this study shows that participants identified pictures, facial images and caricatures more accurately when presented in the same form across measures. Participants were also able to accurately identify images as not seen previously.

Discussion

The results of this study suggest that we identify faces more accurately when we are presented with the same image across all recall conditions. In this study it did not seem to matter whether the person whose face was presented was male or female or whether it was presented as a facial image or caricature. The mean percentages suggest that recall is higher when a face is presented in the same form across all conditions. The numbers also suggest that faces not previously seen are identified as such with a high level of accuracy. These findings do not support the distinctive or caricature hypothesis. This study was designed to re-examine previous research using a computer program that provides a consistent amount of color and detail in both facial images and caricatures. A limitation of this study is that the caricatures used were developed by the researcher and not a trained caricature artist. Future research using professionally developed caricatures with specified levels of exaggeration based on established norms may show different results. The results of this study are hard to generalize because the participants were limited to college sophomores and were not randomly assigned. The stimuli were presented in a controlled and artificial condition. A study using caricatured mug shots in eyewitness identifications may provide results that have external validity.

References

Benson, P. J. , Perrett , D. I. (1991). Perception and recognition of photographic quality facial caricatures: Implications for the recognition of natural images. European Journal of Cognitive Psychology, 3, 105-135.

Mauro, R. , Kubovy, M. (1992). Caricature and face recognition. Memory & Cognition, 20, 433-440.

Rhodes, G. , Brennan, S. , Carey, S. (1987). Identification and ratings of caricatures: Implications the learning. for mental representations of faces. Cognitive Psychology, 19, 473-497.

Rhodes, G. , Moody, J. (1990). Memory representations of unfamiliar faces: Coding of distinctive information. New Zealand Journal of Psychology, 19, 70-78.

Rhodes, G. , Wooding, R. (1989). Laterality effects in identification of caricatures and photographs of famous faces. Brain and Cognition, 9,201-209.

Tversky, B. , Baratz, D. (1985). Memory for faces: Are caricatures better than photographs? Memory & Cognition, 13, 45-49.

Stevenage, S. V. (1995). Demonstration of a caricature advantage in children. Cahiers de Psychologie CognitiveCurrent Psychology of Cognition, 14, 325-341.

Page Last Updated.