Perception of Mooney faces by young infants: The role of local feature visibility, contrast polarity, and motion

https://doi.org/10.1016/j.jecp.2010.10.014Get rights and content

Abstract

We examined the ability of young infants (3- and 4-month-olds) to detect faces in the two-tone images often referred to as Mooney faces. In Experiment 1, this performance was examined in conditions of high and low visibility of local features and with either the presence or absence of the outer head contour. We found that regardless of the presence of the outer head contour, infants preferred upright over inverted two-tone face images only when local features were highly visible (Experiment 1a). We showed that this upright preference disappeared when the contrast polarity of two-tone images was reversed (Experiment 1b), reflecting operation of face-specific mechanisms. In Experiment 2, we investigated whether motion affects infants’ perception of faces in Mooney faces. We found that when the faces appeared to be rigidly moving, infants did show an upright preference in conditions of low visibility of local features (Experiment 2a). Again the preference disappeared when the contrast polarity of the image was reversed (Experiment 2b). Together, these results suggest that young infants have the ability to integrate fragmented image features to perceive faces from two-tone face images, especially if they are moving. This suggests that an interaction between motion and form rather than a purely motion-based process (e.g., structure from motion) facilitates infants’ perception of faces in ambiguous two-tone images.

Introduction

Humans have a remarkable ability to detect and perceive faces; for example, adults can detect facial information even from severely impoverished images such as Mooney faces. The Mooney faces are two-tone images of faces derived by thresholding photographs of faces under asymmetrical lighting conditions (Mooney, 1957). Unlike most natural images, two-tone images do not contain enough cues to differentiate contours arising from illumination effects (e.g., cast shadows, highlights) from contours arising from object structure (e.g., changes in material, pigmentation, or occlusions) without knowing the structure of the object in advance. Although information about local features of faces is severely degraded in Mooney faces and often too ambiguous for recognition as individual features, adults can relatively easily perceive faces when the Mooney faces are presented in an upright orientation (George, Jemel, Fiori, Chaby, & Renault, 2005). Because there are no separable features to be identified, analytical processing is thought to be ineffective with Mooney faces (Latinus & Taylor, 2005) and holistic processing appears to be essential. Moore and Cavanagh (1998) pointed out that typically a two-tone image does not contain cues to disambiguate between contours arising from illumination effects (e.g., cast shadows, highlights) and those arising from object structure (e.g., occlusions, changes in material). They argued that this ambiguity cannot be solved without the guidance of familiarity cues and could block the perceptual recovery of depicted object structure. In fact, Moore and Cavanagh reported that two-tone images of novel objects were poorly recognized, suggesting that interpretation of two-tone images in general requires some kind of top-down knowledge about the structure of the object.

Several research groups have, therefore, used Mooney faces as a tool for investigating the contribution of holistic processing to facial recognition (Farzin et al., 2009, Latinus and Taylor, 2005, Latinus and Taylor, 2006, Le Grand et al., 2006, McKeeff and Tong, 2007). Given that adults readily detect even unfamiliar faces in two-tone Mooney configurations, it is reasonable to assume that this is guided by some general prototypical representation of facial structure that they have developed. In this study, we used two-tone face images to investigate this ability in young infants.

In developmental studies, the ability of infants to detect faces is often assessed by measuring visual preference for upright faces as compared with inverted faces. For example, studies have shown that even newborn infants preferentially look at an upright face or face-like configuration as compared with its inverted counterparts even though the inverted counterparts are complete matches in terms of complexity, amount of energy, and/or any other low-level image properties (Johnson and Morton, 1991, Macchi Cassia et al., 2004, Mondloch et al., 1999, Valenza et al., 1996). Such findings have been interpreted as showing that even newborns have the ability to detect faces. However, there is considerable debate as to whether this preference in newborns reflects face-specific or more general processing mechanisms (for a review, see Simion, Leo, Turati, Valenza, & Dalla Barba, 2007).

For example, Simion and colleagues (Simion et al., 2001, Simion et al., 2002) proposed that the preference for upright faces in the newborns is attributable to a non-face-specific structural property of faces, specifically the “top-heavy” property (i.e., more elements are located in the upper part than in the lower part of the stimuli). Even when using geometrical non-face-like stimuli, Simion and colleagues (2002) found that newborns preferred top-heavy to inverted versions of the same stimuli. Turati, Simion, Milani, and Umiltà (2002) extended this finding by showing that newborns did not prefer face-like to equally top-heavy non-face-like arrangements. Furthermore, newborns preferred top-heavy but non-face-like configurations to face-like configurations in which the position of the elements was shifted toward the bottom half of the stimuli. However, other researchers have shown that newborns’ preference for face is not determined by top-heavy arrangement of facial features alone; it also depends on the presence of ecologically valid contrast polarity relations (e.g., Farroni et al., 2005).

There is also evidence that newborns have a more limited representation of facial structure as compared with older infants (Macchi Cassia et al., 2006, Macchi Cassia et al., 2004, Turati et al., 2005). Using stimuli based on photographs of faces, Macchi Cassia and colleagues (2004) reported that newborns prefer upright over inverted top-heavy scrambled faces, just as they prefer upright over inverted normal faces. In contrast, 3-month-olds did not prefer upright over inverted top-heavy scrambled faces even though they still preferred upright over inverted normal faces. In addition, 3-month-olds, unlike newborns, preferred upright normal faces over upright top-heavy scrambled faces (Macchi Cassia et al., 2006, Turati et al., 2005). This pattern of results suggests that the preference for upright faces becomes more face specific and representations of facial structure become more sophisticated by 3 months of age.

Whereas most of the previous studies used facial photographs or schematic face-like stimuli that contained well-defined features surrounded by a complete contour, some recent studies used images where features and contours were incompletely specified. For example, Gava, Valenza, Turati, and de Schonen (2008) examined the effect of partial occlusion on infants’ detection of faces. They examined preference for upright over inverted faces under two occlusion conditions. In one condition, the eyes and part of the external contours were occluded by vertical bars (High Salience Occlusion condition). In the other condition, the position of the face was shifted relative to the occluding bars so that the mouth, the nose, and a different part of the external contours were occluded (Low Salience Occlusion condition). Gava and colleagues reported that newborn infants preferred the upright faces only in the Low Salience Occlusion condition, where eyes were visible. The results might be interpreted as implying that newborns can detect faces if some low-informative face portions are hidden even if they cannot do so when high salience features are hidden. However, the preference for upright faces observed only in the Low Salience Occlusion condition is consistent with the top-heavy hypothesis because the up–down asymmetry of faces was affected only in the High Salience Occlusion condition.

Unlike the schematic faces and facial photographs used in nearly all previous developmental studies, Mooney face stimuli do not contain individual facial features and the face can be perceived only when the image fragments are processed as a whole. Therefore, as with adults, Mooney faces provide an additional tool for investigating this aspect of face detection ability in infants. Doi, Koga, and Shinohara (2009) first examined Mooney face perception in 6-, 12-, and 18-month-olds. They examined preference for upright over inverted presentations of the same Mooney face images and found that only 18-month-olds showed an upright preference. In contrast, when full grayscale photographs were used as the test stimuli, younger infants also showed a preference for upright faces. These results suggest that the ability to detect and perceive the face in a Mooney face image does not develop until very late in infancy. However, an even more recent study by Leo and Simion (2009) reported that even newborn infants show a preference for upright over inverted Mooney faces.

One way to reconcile these seemingly discrepant findings is to consider that for newborns, with their predominantly low spatial frequency vision, Mooney faces would look more similar to real faces than they would for older infants with better ability to differentiate higher spatial frequency detail. Indeed, Leo and Simion (2009) argued that for newborns Mooney faces would not provide information very different from that provided by photographs of real faces. For infants who have better visual acuity, the spurious edge content present in the two-tone nature of Mooney faces may have interfered with detection.

Another possibility is that Mooney faces are too impoverished for older infants to show upright face preference because they have developed a more sophisticated representation of faces. Previous studies have suggested that more complex and naturalistic facial stimuli were required to elicit a preference in older infants (e.g., Johnson et al., 1992, Mondloch et al., 1999). This possibility suggests that there should be a condition where infants show the upright preference even in the two-tone image provided that the image is rich enough.

Finally, the differential two-tone facial images used in these studies might explain the differential findings. McKone (2004) noted that the visibility of faces differs substantially even within the original Mooney faces (Mooney, 1957). Neither of the previous developmental studies used the original Mooney face images (Mooney, 1957); instead, they used two-tone images produced from a different facial image source. Thus, it could be the case that the Mooney face images used by Leo and Simion (2009) were simply more recognizable than those used by Doi and colleagues (2009).

In this article, we report experiments intended to establish under what conditions, if any, young infants (3- and 4-month-olds) can detect faces in two-tone images. This age group was chosen because previous studies indicate that several abilities relevant to the perception of two-tone Mooney face emerge at around this age. In particular, face detection in two-tone Mooney images is dependent on (a) the ability to integrate image fragments into global structure, (b) sensitivity to shading/shadow cues, and (c) knowledge about facial structure. Findings from several studies provide converging evidence suggesting that these abilities are already present at 3 and 4 months of age. First, studies investigating illusory contour perception show that an ability to perceive coherent global structure from image fragments based on pictorial information alone emerges at 3 or 4 months of age (Ghim, 1990, Kavšek, 2002, Otsuka et al., 2004, Otsuka et al., 2008). Second, some studies suggest that even 3- to 4-month-olds exhibit sensitivity to shading cues and are able to perceive the three-dimensional (3D) structure conveyed by those cues (Bertin and Bhatt, 2006, Bhatt and Waters, 1998). Finally, face recognition studies suggest that infants become sensitive to the basic structure of faces at around this age. Furthermore, and as already mentioned, 3-month-olds discriminate and prefer normal faces over top-heavy scrambled faces (Macchi Cassia et al., 2006, Turati et al., 2005). Furthermore, Bhatt, Bertin, Hayden, and Reed (2005) found that 3-month-olds detected positional change of facial feature when these changes violated the common structure of human faces, so-called first-order relational property (e.g., nose above mouth and two eyes above nose). In contrast, they did not notice the positional change when first-order relational properties were not violated (e.g., a change in the distance between features).

In a pilot study, we used three examples from the original set of Mooney faces (Mooney, 1957) that were judged as easily identifiable by adults. Consistent with recently published data by Doi and colleagues (2009), we found that infants between 3 and 4 months of age showed no upright preference. However, from this finding alone, it was difficult to determine whether the failure was due to (a) the two-tone property of the images, (b) the impoverished nature of the images, or (c) the infants’ inability to integrate fragmented image features to form coherent representations of faces. In the current study, we used parametric variations in the spatial properties of two-tone facial images to differentiate among these alternatives using the upright preference as an index for detection. In Experiment 1a, we varied the luminance threshold and background color to determine the role of isolated features as well as the presence or absence of head outline in face detection in two-tone images. In Experiment 1b, we tested for face specificity of the observed effects by using contrast-reversed two-tone facial images. In Experiment 2a, we investigated the effect of apparent motion on the detection of two-tone face images without isolated facial features. Finally, in Experiment 2b, we examined the face detection by using contrast-reversed version of the apparent motion display.

Section snippets

Experiment 1a: Local facial features and head contour

For this experiment, we created two-tone facial images by thresholding photographs of three female faces by setting all gray-level values above the threshold to white and setting all those below the threshold to black. We chose threshold levels so as to create face configurations that differed with respect to the presence of isolated local facial features (eyes, nose, and mouth). We also manipulated background color so that either it matched hair color and there was no external head contour or

Experiment 1b: Reversed contrast polarity

The aim of Experiment 1b was to establish that the preference for upright two-tone faces shown in Experiment 1a reflects the infants’ face detection. For this purpose, we created contrast-reversed versions of the low-threshold two-tone face images used in Experiment 1a (Fig. 2).

Previous studies have shown that contrast polarity reversal affects face detection in both adults (Tomalski et al., 2009, Torralba and Sinha, 2001) and infants (Dannemiller and Stephens, 1988, Mondloch et al., 1999).

Experiment 2a: Motion vs. static

The aim of Experiment 2 was to test whether motion affects infants’ perception of faces in two-tone Mooney face images that do not contain isolated facial features. Apparent motion displays—displays in which two or more static images presented in rapid succession give the impression of movement—have been used in previous newborn and infant studies investigating the contribution of motion processing to the perception of illusory contours (Kavšek and Yonas, 2006, Valenza and Bulf, 2007),

Experiment 2b: Moving contrast-reversed images

In Experiment 2a, we found that infants preferred upright Mooney faces only in the Moving condition. One possible reason for the better performance in the Moving condition is that the perception of 3D structure of faces is facilitated by motion information (i.e., structure from motion). In fact, a previous study showed that infants as young as 8 weeks could perceive 3D object shapes based solely on motion information (Arterberry & Yonas, 2000). However, to test whether this was critical here, we

General discussion

The current study examined the ability of young infants (3- and 4-month-olds) to perceive faces in two-tone Mooney facial images. Although neither our pilot study using original Mooney faces nor similar face images in Experiment 1a led to an upright face preference, we did find an upright face preference when some isolated features were visible in the images. This indicates somewhat limited face detection from two-tone images in this age group. Any upright preference disappeared when images

Acknowledgments

This study was supported by PRESTO (Japan Science and Technology Agency), by a Grant-in-Aid for Scientific Research on Innovative Areas, “Face Perception and Recognition” from MEXT KAKENHI (20119002), by a Grant-in-Aid for JSPS Fellows (196068) from the Japan Society for the Promotion of Science, and by an Australian Research Council Discovery Grant (DP1096099). We thank H. Ichikawa, J. Yang, M. Kobayashi, A. Tsuruhara, E. Nakato, W. Yamashita, and Y. Yamazaki for their help in collecting the

References (53)

  • C. Moore et al.

    Recovery of 3D volume from 2-tone images of novel objects

    Cognition

    (1998)
  • Y. Otsuka et al.

    Infants’ perception of illusory contours in static and moving figures

    Journal of Experimental Child Psychology

    (2003)
  • G. Schwarzer et al.

    Face processing in 8-month-old infants: Evidence for configural and analytical processing

    Vision Research

    (2003)
  • C. Turati et al.

    Three-month-old visual preference for faces and its underlying visual processing mechanisms

    Journal of Experimental Child Psychology

    (2005)
  • S.M. Anstis

    Homage to Pete Thompson: The Tony Blair illusion

    Perception

    (2005)
  • M.E. Arterberry et al.

    Perception of three-dimensional shape specified by optic flow by 8-week-old infants

    Perception & Psychophysics

    (2000)
  • R.S. Bhatt et al.

    Face processing in infancy: Developmental changes in the use of different kinds of relational information

    Child Development

    (2005)
  • H. Bulf et al.

    The role of rigid motion in newborns’ face recognition

    Visual Cognition

    (2010)
  • P. Cavanagh et al.

    Shape from shadows

    Journal of Experimental Psychology: Human Perception and Performance

    (1989)
  • J.L. Dannemiller et al.

    A critical test of infant pattern preference models

    Child Development

    (1988)
  • T. Farroni et al.

    Newborns’ preference for face-relevant stimuli: Effects of contrast polarity

    Proceedings of the National Academy of Sciences of the United States of America

    (2005)
  • F. Farzin et al.

    Holistic crowding of Mooney faces

    Journal of Vision

    (2009)
  • L. Gava et al.

    Effect of partial occlusion on newborns’ face preference and recognition

    Developmental Science

    (2008)
  • M.H. Johnson et al.

    Biology and cognitive development: The case of face recognition

    (1991)
  • M.J. Kavšek

    The perception of static subjective contours in infancy

    Child Development

    (2002)
  • M.J. Kavšek et al.

    The perception of moving subjective contours by 4 month old infants

    Perception

    (2006)
  • Cited by (0)

    View full text