- Email: [email protected]
Orientation-Contingent Face Aftereffects and Implications for Face-Coding Mechanisms
Current Biology, Vol. 14, 2119–2123, December 14, 2004, ©2004 Elsevier Ltd. All rights reserved.
DOI 10.1016/j .c ub . 20 04 . 11 .0 5 3
Orientation-Contingent Face Aftereffects and Implications for Face-Coding Mechanisms Gillian Rhodes,1,* Linda Jeffery,1 Tamara L. Watson,2 Emma Jaquet,1 Chris Winkler,1 and Colin W.G. Clifford 2 1 University of Western Australia Stirling Highway Crawley, Perth WA 6009 Australia 2 The University of Sydney NSW 2006 Australia
Summary Humans have an impressive ability to discriminate between faces despite their similarity as visual patterns [1, 2]. This expertise relies on configural coding of spatial relations between face features and/or holistic coding of overall facial structure [2–6]. These expert face-coding mechanisms appear to be engaged most effectively by upright faces, with inverted faces engaging primarily feature-coding mechanisms [2, 7–11]. We show that opposite figural aftereffects can be induced simultaneously for upright and inverted faces, demonstrating that distinct neural populations code upright and inverted faces. This result also suggests that expert (upright) face-coding mechanisms can be selectively adapted. These aftereffects occur for judgments of face normality and face gender and are robust to changes in face size, ruling out adaptation of lowlevel, retinotopically organized coding mechanisms. Our results suggest a resolution of a paradox in the face recognition literature. Neuroimaging studies have found surprisingly little orientation selectivity in the fusiform face area (FFA) despite evidence that this region plays a role in expert face coding [12–14] and that expert face-coding mechanisms are selectively engaged by upright faces [2, 7–11]. Our results, demonstrating orientation-contingent adaptation of facecoding mechanisms, suggest that the FFA’s apparent lack of orientation selectivity may be an artifact of averaging across distinct populations within the FFA that respond to upright and inverted faces. Results and Discussion In Experiment 1, we attempted to simultaneously induce opposite figural aftereffects in upright and inverted faces by adapting people to upright and inverted faces with opposite distortions (expanded or contracted) (Figure 1A). If distinct high-level mechanisms code upright and inverted faces, then we should see opposite aftereffects in upright and inverted faces (orientation-contingent aftereffects). If not, then the opposite adapting distortions should cancel, and no aftereffects should be *Correspondence: [email protected]
seen. This assumption requires that upright and inverted faces match on low-level properties, and although this is true for many low-level properties (spatial frequency, contrast, etc.), some differences do remain (e.g., in the locations of edges). Therefore, we attempted to rule out any contribution of low-level adaptation by introducing a size change between adapting and test faces (Experiment 1B). We sought converging evidence that any observed orientation-contingent aftereffects reflect adaptation of face-coding mechanisms in a second experiment in which we attempted to induce orientation-contingent aftereffects in the perception of face gender (Figure 2). Although not exclusive to faces (bodies also have a gender), gender is not an attribute possessed by most objects. Therefore, orientation-contingent aftereffects in the perception of face gender would be consistent with adaptation of high-level face-coding mechanisms rather than more generic object-coding mechanisms.
Experiment 1A We attempted to induce simultaneous, opposite figural aftereffects in upright and inverted faces. Forty undergraduates (13 male) adapted to six upright faces and six different inverted faces [exposure (exp) ⫽ 1000 ms, interstimulus interval (ISI) ⫽ 200 ms] sampled randomly for 2 min. Participants saw either upright faces with “contracted” internal features and inverted faces with “expanded” features (n ⫽ 20) or vice versa (n ⫽ 20) (see Figure 1 for details). Before and after adapting, participants rated the perceived normality (1 ⫽ low, 9 ⫽ high) of 136 upright (8 faces ⫻ 17 distortion levels, Figure 1A) and 136 inverted test faces shown in random order. The test faces were different identities from the adapting faces. To maintain adaptation during testing, three upright and three inverted adapting faces were shown in random order (exp ⫽ 1000 ms, ISI ⫽ 200 ms) before each test face (exp ⫽ 1000 ms, ISI ⫽ 400 ms). Faces were viewed from 57cm and subtended approximately 9.6 ⫻ 12.7 degrees. Simultaneous, opposite aftereffects occurred for upright and inverted faces, with the most normal-looking distortion shifting toward the adapting distortion in each case (Figure 1B). A two-way ANOVA with orientation as a repeated measures factor and adapting condition (upright contracted, inverted expanded [UCIE] or upright expanded, inverted contracted [UEIC]) as a betweensubjects factor was conducted on the aftereffect, which was measured as the shift in distortion level that looked most normal. A significant main effect of adapting condition, F (1, 34) ⫽ 31.51, p ⬍ 0.001, was qualified by a significant interaction with test orientation, F (1, 34) ⫽ 149.12, p ⬍ 0.001, indicating orientation-contingent aftereffects. Planned t tests confirmed that significant aftereffects occurred for both upright (UCIE: t (18) ⫽ 4.05, p ⬍ .01, UEIC: t (16) ⫽ 2.74, p ⬍ .02) and inverted (UCIE: t (18) ⫽ 11.00, p ⬍ .001, UEIC: t (16) ⫽ 8.68, p ⬍ .001) faces. (One outlier was dropped from the UCIE condition
Current Biology 2120
Figure 1. Test Faces and Results for Experiments 1A and 1B (A) A test face, shown at eight distortion levels, created using the spherize function in Adobe Photoshop [15]. In the negative (contracted) distortions, internal features contract and move closer together, whereas in positive (expanded) distortions, they expand and move apart. Seventeen distortion levels were actually used in the test phase (⫺70, ⫺50, ⫺30, ⫺20, ⫺15, ⫺10, ⫺5, ⫺2, 0, ⫹2, ⫹5, ⫹10, ⫹15,⫹ 20, ⫹30, ⫹50, ⫹70). The adapting distortions were ⫺50 (contracted) and ⫹70 (expanded), which looked equally distorted (for both upright and inverted faces) in pilot testing with 12 participants. (B) Experiment 1A: The aftereffect (mean ⫾ SE) for upright and inverted test faces in each adapting condition (upright faces contracted, inverted faces expanded; upright faces expanded, inverted faces contracted). For each participant, the aftereffect was calculated by subtracting the most normal-looking distortion level after adapting from the most normal-looking distortion level before adapting, for each condition. The most normal-looking distortion levels were obtained by fitting third-order polynomials to each participant’s normality ratings plotted as a function of distortion level and calculating their maxima, for each condition. Fits were good (mean ⫽ 0.86, SD ⫽ 0.08 minimum ⫽ 0.53; 12 raters were replaced because not all of their fits exceeded .50). (C) Experiment 1B: The aftereffect (mean ⫾ SE), calculated as above, for upright and inverted test faces of different sizes. Third-order polynomial fits were good (mean ⫽ 0.88, SD ⫽ 0.06, minimum ⫽ 0.70; 3 raters were replaced because not all of their fits exceeded .70).
Face Aftereffects 2121
Figure 2. Adapting Faces, Test Faces, and Results for Experiment 2 (A) Examples of the male and female faces used as adapting stimuli in Experiment 2. (B) Test stimuli were morphs drawn from a continuum ranging from an average male face to an average female face (averages made from 24 male and 24 female faces respectively). (C) Averaged data (⫾SE) from six subjects, showing the position of the subjective category boundary between male and female for upright and inverted test faces, for three different adapting conditions.
because the pre-post-shift was more than two standard deviations from the mean. Three outliers were dropped from the UEIC condition.) Unexpectedly, the aftereffect was greater for inverted than upright faces, for both, t ⬎ 2.86 and p ⬍ .008. The adapting distortions used for upright and inverted faces were chosen to look equally distorted based on pilot testing, so the difference cannot be due to differences in the perceived level of adapting distortion. A stronger aftereffect for inverted faces may indicate greater plasticity in the processing of less-common stimuli. As is typical for contingent aftereffects, those observed here were smaller than the simple aftereffects (without opposite distortions presented during adaptation) obtained with these stimuli under similar conditions [15]. This reduction is expected because upright and inverted faces match on many low-level attributes (spatial frequency, contrast, etc) so that the effect
of opposite distortions would cancel, thereby reducing the contribution of low-level adaptation to the aftereffects. Experiment 1B This experiment was identical to Experiment 1A except that the adaptation and test images differed in size (12.1 ⫻ 15.6 degrees and 5.3 ⫻ 7.0 degrees, respectively), and a chin rest was used for ensuring a constant viewing distance. Twenty-three undergraduates (4 male) participated. Opposite aftereffects were again observed for upright and inverted faces with the most normallooking distortion shifting toward the adapting distortion in each case (Figure 1C). As in Experiment 1A, adapting condition interacted with test orientation, F (1, 21) ⫽ 58.73, p ⬍ 0.001, and planned t tests confirmed that significant aftereffects occurred for both upright [UCIE:
Current Biology 2122
t (11) ⫽ 3.96, p ⬍ 0.003, UEIC: t (10) ⫽ 3.44, p ⬍ 0.007, and inverted faces, UCIE: t (11) ⫽ 3.74, p ⬍ 0.004; UEIC: t(10) ⫽ 3.97, p ⬍ 0.004]. Therefore, the orientation-contingent aftereffects were robust to a size change, ruling out a contribution from adaptation of low-level, retinotopically organized coding mechanisms. No other effects were significant, and we did not replicate the larger aftereffects for inverted faces found in Experiment 1A, for both, t ⬍ 1. Comparing the size of the aftereffects in Experiments 1A and 1B, the introduction of a size change appears to have had a minimal impact on the aftereffect for upright faces. However, for inverted faces, the aftereffect appeared reduced in the second experiment. These differences suggest that adaptation of low-level mechanisms may contribute more to aftereffects for inverted faces than for upright faces.
Experiment 2 We attempted to induce simultaneous, opposite aftereffects in the perceived gender of upright and inverted faces. In contrast to Experiment 1, we used an adaptive psychophysical procedure [16] and a within-participants design. Six subjects (3 male) were adapted for 2 min to alternating upright and inverted faces (exp ⫽ 1000 ms) sampled randomly from eight upright faces and eight different inverted faces. Participants saw either upright male faces and inverted female faces or vice versa (Figure 2A). Before and after adapting, participants completed two sets of 120 trials in which they were required to make a forced-choice judgment of the gender of a face selected from a continuum ranging from an average male face to an average female face (Figure 2B). Each set of 120 trials consisted of four interleaved, adaptive psychophysical procedures [16], which, in 30 trials, provided an estimate of the point along the gender continuum at which the subject was equally likely to respond male as female: the subjective gender category boundary. To maintain adaptation during testing, three upright (U) and three inverted (I) adapting faces were randomly alternated (UIUIUI or IUIUIU; exp ⫽ 1000 ms) before each test face (exp ⫽ 1000 ms, ISI ⫽ 400 ms). A chin rest was used to ensure a constant viewing distance. Adapting faces subtended 11 ⫻ 15 degrees. Test faces were smaller, subtending 5 ⫻ 7 degrees. Each subject completed two sets of 120 trials for one combination of adapting orientation and gender, followed, 72 hr later, by two more sets with the opposite adapting combination. Thus, for each subject, four estimates of the subjective gender category boundary were obtained for each pairing of adapting combination (male upright female inverted, unadapted, and female upright male inverted) and test orientation (upright, inverted). A two-way ANOVA with adapting condition and test orientation as repeated-measure factors was conducted on the gender category boundary estimates. Adapting condition interacted with test orientation, F (2, 10) ⫽ 4.87, p ⬍ .03, demonstrating orientation-contingent aftereffects for perceived gender (Figure 2). Curiously, there was also a significant main effect of test orientation, F (1, 5)⫽ 11.15, p ⬍ .02 with inverted faces appearing more female than upright faces.
In summary, we induced simultaneous, opposite aftereffects in upright and inverted faces, indicating that upright and inverted faces are coded by distinct neural populations. Previous studies have shown partial transfer of figural aftereffects between upright and inverted faces [17, 18], suggesting that the orientation selectivity is not perfect. The aftereffects reported here were robust to changes in face size, ruling out adaptation of lowlevel, retinotopically organized coding mechanisms. They are consistent with adaptation of face-coding mechanisms because they affected perceptions of gender, an attribute not generally possessed by objects or shapes, and normality, which requires reference to information about the statistics of face populations. The adapting distortions used here altered both configural information and feature appearance and, therefore, had the potential to adapt both configural and feature-coding mechanisms. We speculate that the aftereffects observed for upright and inverted faces may reflect selective adaptation of configural and feature-coding mechanisms, respectively. Our results suggest a potential resolution of a paradox in the face recognition literature. Despite behavioral evidence that upright, but not inverted, faces engage expert face-coding mechanisms, the fusiform face area (FFA) responds almost as strongly to inverted as to upright faces [19, 20]. This lack of orientation specificity is difficult to reconcile with results that implicate the FFA in expert face processing. The FFA responds more strongly to faces than other objects [12, 13], is sensitive to individuating information in faces [21, 22], is more active (in the right hemisphere) when matching whole faces than face parts [23], and responds more strongly to own-race faces than to other-race faces, with which participants have less expertise [14]. We suggest that the limited orientation specificity of the FFA may result from averaging activity across distinct populations within the FFA that respond to upright and inverted faces. Brain-imaging studies will be needed to test this hypothesis. Functional magnetic resonance (fMR) adaptation [24] could determine whether neurons in the FFA, and other face-coding regions, are orientation selective by examining whether their responses recover from habituation/adaptation when face orientation changes from upright to inverted, or vice versa. fMR adaptation could also be used for determining which areas contain neurons that code configural and feature information in faces. Although the precise neural locus of the orientation-contingent face aftereffects reported here remains to be determined, these aftereffects show that adaptation can be a powerful tool for exploring the computational mechanisms of face coding [18, 25–28]. Acknowledgments This work was supported by the Australian Research Council. We thank Concetta Morrone, David Burr, and Elinor McKone for helpful discussions about this work. We also thank Glyn Cowe for morphing expertise and Ben Sachtler for technical assistance. Received: March 2, 2004 Revised: October 4, 2004 Accepted: October 4, 2004 Published: December 14, 2004
Face Aftereffects 2123
References 1. Bharick, H.P., Bharick, P.O., and Wittlinger, R.P. (1975). Fifty years of memory for names and faces: A cross-sectional approach. J. Exp. Psychol. Gen. 104, 54–57. 2. Peterson, M.A., and Rhodes, G. eds. (2003). Perception of faces, objects, and scenes: Analytic and holistic processes (London: Oxford University Press). 3. Maurer, D., Le Grand, R., and Mondloch, C.J. (2002). The many faces of configural processing. Trends. Cogn. Sci. 6, 255–260. 4. Murray, J.E., Rhodes, G., and Schuchinsky, M. (2003). When is a face not a face? The effects of misorientation on mechanisms of face perception. In Perception of faces, objects, and scenes: Analytic and holistic processes, M.A. Peterson and G. Rhodes, eds. (London: Oxford University Press), pp. 75–117. 5. Rhodes, G., Brake, S., Taylor, K., and Tan, S. (1989). Expertise and configural coding in face recognition. Br. J. Psychol. 80, 313–331. 6. Tanaka, J.W., Kiefer, M., and Bukach, C.M. (2004). A holistic account of the own-race effect in face recognition: evidence from a cross-cultural study. Cognition 93, B1–B9. 7. Yin, R.K. (1969). Looking at upside-down faces. J. Exp. Psychol. 81, 141–145. 8. Moscovitch, M., Winocur, G., and Behrmann, M. (1997). What is special about face recognition? Nineteen experiments on a person with visual object agnosia and dyslexia but normal face recognition. J. Cogn. Neurosci. 9, 555–604. 9. Bartlett, J.C., and Searcy, J. (1993). Inversion and configuration of faces. Cognit. Psychol. 25, 281–316. 10. Diamond, R., and Carey, S. (1986). Why faces are and are not special: An effect of expertise. J. Exp. Psychol. Gen. 115, 107–117. 11. Pellicano, E., and Rhodes, G. (2003). Holistic processing of faces in preschool children and adults. Psychol. Sci. 14, 618–622. 12. Kanwisher, N., McDermott, J., and Chun, M.M. (1997). The fusiform face area: A module in human extrastriate cortex specialized for face perception. J. Neurosci. 17, 4302–4311. 13. Kanwisher, N. (2000). Domain specificity in face perception. Nat. Neurosci. 3, 759–763. 14. Golby, A.J., Gabrieli, J.D.E., Chiao, J.Y., and Eberhart, J. (2001). Differential responses in the fusiform region to same-race and other-race faces. Nat. Neurosci. 4, 845–850. 15. Rhodes, G., Jeffery, L., Watson, T.L., Clifford, C.W.G., and Nakayama, K. (2003). Fitting the mind to the world: Face adaptation and attractiveness aftereffects. Psychol. Sci. 14, 558–566. 16. Kontsevich, L.L., and Tyler, C.W. (1999). Bayesian adaptive estimation of psychometric slope and threshold. Vision Res. 39, 2729–2737. 17. Watson, T.L., and Clifford, C.W.G. (2003). Pulling faces: An investigation of the face-distortion aftereffect. Perception 32, 1109–1116. 18. Webster, M.A., and MacLin, O.H. (1999). Figural aftereffects in the perception of faces. Psychon. Bull. Rev. 6, 647–653. 19. Haxby, J.V., Hoffman, E.A., and Gobbini, M. (2002). Human neural systems for face recognition and social communication. Biol. Psychiatry 51, 59–67. 20. Rossion, B., and Gauthier, I. (2002). How does the brain process upright and inverted faces? Behav. Cogn. Neurosci. Rev. 1, 63–75. 21. George, N., Dolan, R.J., Fink, G.R., Baylis, G.C., Russell, C., and Driver, J. (1999). Contrast polarity and face recognition in the human fusiform gyrus. Nat. Neurosci. 2, 574–580. 22. Gauthier, I., Tarr, M.J., Moylan, J., Skudlarski, P., Gore, J.C., and Anderson, A.W. (2000). The fusiform “face area” is part of a network that processes faces at the individual level. J. Cogn. Neurosci. 12, 495–504. 23. Rossion, B., Dricot, L., Devolder, A., Bodart, J.-M., Crommelinck, M., de Gelder, B., and Zoontjes, R. (2000). Hemispheric asymmetries for whole-based and part-based face processing in the human fusiform gyrus. J. Cogn. Neurosci. 12, 793–802. 24. Kourtzi, Z., and Grill-Spector, K. (2004). fMRI adaptation: a tool for studying visual representations in the primate brain. In Fitting the Mind to the World: Adaptation and Aftereffects in High-
25.
26. 27.
28.
Level Vision., G. Rhodes and C. Clifford, eds. (Oxford: Oxford University Press), in press. Clifford, C.W.G., and Rhodes, G. eds. (2004). Fitting the Mind to the World: Adaptation and Aftereffects in High-Level Vision (Oxford: Oxford University Press), in press. Hurlbert, A. (2001). Trading faces. Nat. Neurosci. 4, 3–5. Leopold, D.A., O’Toole, A.J., Vetter, T., and Blanz, V. (2001). Prototype-referenced shape encoding revealed by high-level aftereffects. Nat. Neurosci. 4, 89–94. Webster, M.A., Kaping, D., Mizokami, Y., and Duhamel, P. (2004). Adaptation to natural face categories. Nature 428, 557–561.
DOI 10.1016/j .c ub . 20 04 . 11 .0 5 3
Orientation-Contingent Face Aftereffects and Implications for Face-Coding Mechanisms Gillian Rhodes,1,* Linda Jeffery,1 Tamara L. Watson,2 Emma Jaquet,1 Chris Winkler,1 and Colin W.G. Clifford 2 1 University of Western Australia Stirling Highway Crawley, Perth WA 6009 Australia 2 The University of Sydney NSW 2006 Australia
Summary Humans have an impressive ability to discriminate between faces despite their similarity as visual patterns [1, 2]. This expertise relies on configural coding of spatial relations between face features and/or holistic coding of overall facial structure [2–6]. These expert face-coding mechanisms appear to be engaged most effectively by upright faces, with inverted faces engaging primarily feature-coding mechanisms [2, 7–11]. We show that opposite figural aftereffects can be induced simultaneously for upright and inverted faces, demonstrating that distinct neural populations code upright and inverted faces. This result also suggests that expert (upright) face-coding mechanisms can be selectively adapted. These aftereffects occur for judgments of face normality and face gender and are robust to changes in face size, ruling out adaptation of lowlevel, retinotopically organized coding mechanisms. Our results suggest a resolution of a paradox in the face recognition literature. Neuroimaging studies have found surprisingly little orientation selectivity in the fusiform face area (FFA) despite evidence that this region plays a role in expert face coding [12–14] and that expert face-coding mechanisms are selectively engaged by upright faces [2, 7–11]. Our results, demonstrating orientation-contingent adaptation of facecoding mechanisms, suggest that the FFA’s apparent lack of orientation selectivity may be an artifact of averaging across distinct populations within the FFA that respond to upright and inverted faces. Results and Discussion In Experiment 1, we attempted to simultaneously induce opposite figural aftereffects in upright and inverted faces by adapting people to upright and inverted faces with opposite distortions (expanded or contracted) (Figure 1A). If distinct high-level mechanisms code upright and inverted faces, then we should see opposite aftereffects in upright and inverted faces (orientation-contingent aftereffects). If not, then the opposite adapting distortions should cancel, and no aftereffects should be *Correspondence: [email protected]
seen. This assumption requires that upright and inverted faces match on low-level properties, and although this is true for many low-level properties (spatial frequency, contrast, etc.), some differences do remain (e.g., in the locations of edges). Therefore, we attempted to rule out any contribution of low-level adaptation by introducing a size change between adapting and test faces (Experiment 1B). We sought converging evidence that any observed orientation-contingent aftereffects reflect adaptation of face-coding mechanisms in a second experiment in which we attempted to induce orientation-contingent aftereffects in the perception of face gender (Figure 2). Although not exclusive to faces (bodies also have a gender), gender is not an attribute possessed by most objects. Therefore, orientation-contingent aftereffects in the perception of face gender would be consistent with adaptation of high-level face-coding mechanisms rather than more generic object-coding mechanisms.
Experiment 1A We attempted to induce simultaneous, opposite figural aftereffects in upright and inverted faces. Forty undergraduates (13 male) adapted to six upright faces and six different inverted faces [exposure (exp) ⫽ 1000 ms, interstimulus interval (ISI) ⫽ 200 ms] sampled randomly for 2 min. Participants saw either upright faces with “contracted” internal features and inverted faces with “expanded” features (n ⫽ 20) or vice versa (n ⫽ 20) (see Figure 1 for details). Before and after adapting, participants rated the perceived normality (1 ⫽ low, 9 ⫽ high) of 136 upright (8 faces ⫻ 17 distortion levels, Figure 1A) and 136 inverted test faces shown in random order. The test faces were different identities from the adapting faces. To maintain adaptation during testing, three upright and three inverted adapting faces were shown in random order (exp ⫽ 1000 ms, ISI ⫽ 200 ms) before each test face (exp ⫽ 1000 ms, ISI ⫽ 400 ms). Faces were viewed from 57cm and subtended approximately 9.6 ⫻ 12.7 degrees. Simultaneous, opposite aftereffects occurred for upright and inverted faces, with the most normal-looking distortion shifting toward the adapting distortion in each case (Figure 1B). A two-way ANOVA with orientation as a repeated measures factor and adapting condition (upright contracted, inverted expanded [UCIE] or upright expanded, inverted contracted [UEIC]) as a betweensubjects factor was conducted on the aftereffect, which was measured as the shift in distortion level that looked most normal. A significant main effect of adapting condition, F (1, 34) ⫽ 31.51, p ⬍ 0.001, was qualified by a significant interaction with test orientation, F (1, 34) ⫽ 149.12, p ⬍ 0.001, indicating orientation-contingent aftereffects. Planned t tests confirmed that significant aftereffects occurred for both upright (UCIE: t (18) ⫽ 4.05, p ⬍ .01, UEIC: t (16) ⫽ 2.74, p ⬍ .02) and inverted (UCIE: t (18) ⫽ 11.00, p ⬍ .001, UEIC: t (16) ⫽ 8.68, p ⬍ .001) faces. (One outlier was dropped from the UCIE condition
Current Biology 2120
Figure 1. Test Faces and Results for Experiments 1A and 1B (A) A test face, shown at eight distortion levels, created using the spherize function in Adobe Photoshop [15]. In the negative (contracted) distortions, internal features contract and move closer together, whereas in positive (expanded) distortions, they expand and move apart. Seventeen distortion levels were actually used in the test phase (⫺70, ⫺50, ⫺30, ⫺20, ⫺15, ⫺10, ⫺5, ⫺2, 0, ⫹2, ⫹5, ⫹10, ⫹15,⫹ 20, ⫹30, ⫹50, ⫹70). The adapting distortions were ⫺50 (contracted) and ⫹70 (expanded), which looked equally distorted (for both upright and inverted faces) in pilot testing with 12 participants. (B) Experiment 1A: The aftereffect (mean ⫾ SE) for upright and inverted test faces in each adapting condition (upright faces contracted, inverted faces expanded; upright faces expanded, inverted faces contracted). For each participant, the aftereffect was calculated by subtracting the most normal-looking distortion level after adapting from the most normal-looking distortion level before adapting, for each condition. The most normal-looking distortion levels were obtained by fitting third-order polynomials to each participant’s normality ratings plotted as a function of distortion level and calculating their maxima, for each condition. Fits were good (mean ⫽ 0.86, SD ⫽ 0.08 minimum ⫽ 0.53; 12 raters were replaced because not all of their fits exceeded .50). (C) Experiment 1B: The aftereffect (mean ⫾ SE), calculated as above, for upright and inverted test faces of different sizes. Third-order polynomial fits were good (mean ⫽ 0.88, SD ⫽ 0.06, minimum ⫽ 0.70; 3 raters were replaced because not all of their fits exceeded .70).
Face Aftereffects 2121
Figure 2. Adapting Faces, Test Faces, and Results for Experiment 2 (A) Examples of the male and female faces used as adapting stimuli in Experiment 2. (B) Test stimuli were morphs drawn from a continuum ranging from an average male face to an average female face (averages made from 24 male and 24 female faces respectively). (C) Averaged data (⫾SE) from six subjects, showing the position of the subjective category boundary between male and female for upright and inverted test faces, for three different adapting conditions.
because the pre-post-shift was more than two standard deviations from the mean. Three outliers were dropped from the UEIC condition.) Unexpectedly, the aftereffect was greater for inverted than upright faces, for both, t ⬎ 2.86 and p ⬍ .008. The adapting distortions used for upright and inverted faces were chosen to look equally distorted based on pilot testing, so the difference cannot be due to differences in the perceived level of adapting distortion. A stronger aftereffect for inverted faces may indicate greater plasticity in the processing of less-common stimuli. As is typical for contingent aftereffects, those observed here were smaller than the simple aftereffects (without opposite distortions presented during adaptation) obtained with these stimuli under similar conditions [15]. This reduction is expected because upright and inverted faces match on many low-level attributes (spatial frequency, contrast, etc) so that the effect
of opposite distortions would cancel, thereby reducing the contribution of low-level adaptation to the aftereffects. Experiment 1B This experiment was identical to Experiment 1A except that the adaptation and test images differed in size (12.1 ⫻ 15.6 degrees and 5.3 ⫻ 7.0 degrees, respectively), and a chin rest was used for ensuring a constant viewing distance. Twenty-three undergraduates (4 male) participated. Opposite aftereffects were again observed for upright and inverted faces with the most normallooking distortion shifting toward the adapting distortion in each case (Figure 1C). As in Experiment 1A, adapting condition interacted with test orientation, F (1, 21) ⫽ 58.73, p ⬍ 0.001, and planned t tests confirmed that significant aftereffects occurred for both upright [UCIE:
Current Biology 2122
t (11) ⫽ 3.96, p ⬍ 0.003, UEIC: t (10) ⫽ 3.44, p ⬍ 0.007, and inverted faces, UCIE: t (11) ⫽ 3.74, p ⬍ 0.004; UEIC: t(10) ⫽ 3.97, p ⬍ 0.004]. Therefore, the orientation-contingent aftereffects were robust to a size change, ruling out a contribution from adaptation of low-level, retinotopically organized coding mechanisms. No other effects were significant, and we did not replicate the larger aftereffects for inverted faces found in Experiment 1A, for both, t ⬍ 1. Comparing the size of the aftereffects in Experiments 1A and 1B, the introduction of a size change appears to have had a minimal impact on the aftereffect for upright faces. However, for inverted faces, the aftereffect appeared reduced in the second experiment. These differences suggest that adaptation of low-level mechanisms may contribute more to aftereffects for inverted faces than for upright faces.
Experiment 2 We attempted to induce simultaneous, opposite aftereffects in the perceived gender of upright and inverted faces. In contrast to Experiment 1, we used an adaptive psychophysical procedure [16] and a within-participants design. Six subjects (3 male) were adapted for 2 min to alternating upright and inverted faces (exp ⫽ 1000 ms) sampled randomly from eight upright faces and eight different inverted faces. Participants saw either upright male faces and inverted female faces or vice versa (Figure 2A). Before and after adapting, participants completed two sets of 120 trials in which they were required to make a forced-choice judgment of the gender of a face selected from a continuum ranging from an average male face to an average female face (Figure 2B). Each set of 120 trials consisted of four interleaved, adaptive psychophysical procedures [16], which, in 30 trials, provided an estimate of the point along the gender continuum at which the subject was equally likely to respond male as female: the subjective gender category boundary. To maintain adaptation during testing, three upright (U) and three inverted (I) adapting faces were randomly alternated (UIUIUI or IUIUIU; exp ⫽ 1000 ms) before each test face (exp ⫽ 1000 ms, ISI ⫽ 400 ms). A chin rest was used to ensure a constant viewing distance. Adapting faces subtended 11 ⫻ 15 degrees. Test faces were smaller, subtending 5 ⫻ 7 degrees. Each subject completed two sets of 120 trials for one combination of adapting orientation and gender, followed, 72 hr later, by two more sets with the opposite adapting combination. Thus, for each subject, four estimates of the subjective gender category boundary were obtained for each pairing of adapting combination (male upright female inverted, unadapted, and female upright male inverted) and test orientation (upright, inverted). A two-way ANOVA with adapting condition and test orientation as repeated-measure factors was conducted on the gender category boundary estimates. Adapting condition interacted with test orientation, F (2, 10) ⫽ 4.87, p ⬍ .03, demonstrating orientation-contingent aftereffects for perceived gender (Figure 2). Curiously, there was also a significant main effect of test orientation, F (1, 5)⫽ 11.15, p ⬍ .02 with inverted faces appearing more female than upright faces.
In summary, we induced simultaneous, opposite aftereffects in upright and inverted faces, indicating that upright and inverted faces are coded by distinct neural populations. Previous studies have shown partial transfer of figural aftereffects between upright and inverted faces [17, 18], suggesting that the orientation selectivity is not perfect. The aftereffects reported here were robust to changes in face size, ruling out adaptation of lowlevel, retinotopically organized coding mechanisms. They are consistent with adaptation of face-coding mechanisms because they affected perceptions of gender, an attribute not generally possessed by objects or shapes, and normality, which requires reference to information about the statistics of face populations. The adapting distortions used here altered both configural information and feature appearance and, therefore, had the potential to adapt both configural and feature-coding mechanisms. We speculate that the aftereffects observed for upright and inverted faces may reflect selective adaptation of configural and feature-coding mechanisms, respectively. Our results suggest a potential resolution of a paradox in the face recognition literature. Despite behavioral evidence that upright, but not inverted, faces engage expert face-coding mechanisms, the fusiform face area (FFA) responds almost as strongly to inverted as to upright faces [19, 20]. This lack of orientation specificity is difficult to reconcile with results that implicate the FFA in expert face processing. The FFA responds more strongly to faces than other objects [12, 13], is sensitive to individuating information in faces [21, 22], is more active (in the right hemisphere) when matching whole faces than face parts [23], and responds more strongly to own-race faces than to other-race faces, with which participants have less expertise [14]. We suggest that the limited orientation specificity of the FFA may result from averaging activity across distinct populations within the FFA that respond to upright and inverted faces. Brain-imaging studies will be needed to test this hypothesis. Functional magnetic resonance (fMR) adaptation [24] could determine whether neurons in the FFA, and other face-coding regions, are orientation selective by examining whether their responses recover from habituation/adaptation when face orientation changes from upright to inverted, or vice versa. fMR adaptation could also be used for determining which areas contain neurons that code configural and feature information in faces. Although the precise neural locus of the orientation-contingent face aftereffects reported here remains to be determined, these aftereffects show that adaptation can be a powerful tool for exploring the computational mechanisms of face coding [18, 25–28]. Acknowledgments This work was supported by the Australian Research Council. We thank Concetta Morrone, David Burr, and Elinor McKone for helpful discussions about this work. We also thank Glyn Cowe for morphing expertise and Ben Sachtler for technical assistance. Received: March 2, 2004 Revised: October 4, 2004 Accepted: October 4, 2004 Published: December 14, 2004
Face Aftereffects 2123
References 1. Bharick, H.P., Bharick, P.O., and Wittlinger, R.P. (1975). Fifty years of memory for names and faces: A cross-sectional approach. J. Exp. Psychol. Gen. 104, 54–57. 2. Peterson, M.A., and Rhodes, G. eds. (2003). Perception of faces, objects, and scenes: Analytic and holistic processes (London: Oxford University Press). 3. Maurer, D., Le Grand, R., and Mondloch, C.J. (2002). The many faces of configural processing. Trends. Cogn. Sci. 6, 255–260. 4. Murray, J.E., Rhodes, G., and Schuchinsky, M. (2003). When is a face not a face? The effects of misorientation on mechanisms of face perception. In Perception of faces, objects, and scenes: Analytic and holistic processes, M.A. Peterson and G. Rhodes, eds. (London: Oxford University Press), pp. 75–117. 5. Rhodes, G., Brake, S., Taylor, K., and Tan, S. (1989). Expertise and configural coding in face recognition. Br. J. Psychol. 80, 313–331. 6. Tanaka, J.W., Kiefer, M., and Bukach, C.M. (2004). A holistic account of the own-race effect in face recognition: evidence from a cross-cultural study. Cognition 93, B1–B9. 7. Yin, R.K. (1969). Looking at upside-down faces. J. Exp. Psychol. 81, 141–145. 8. Moscovitch, M., Winocur, G., and Behrmann, M. (1997). What is special about face recognition? Nineteen experiments on a person with visual object agnosia and dyslexia but normal face recognition. J. Cogn. Neurosci. 9, 555–604. 9. Bartlett, J.C., and Searcy, J. (1993). Inversion and configuration of faces. Cognit. Psychol. 25, 281–316. 10. Diamond, R., and Carey, S. (1986). Why faces are and are not special: An effect of expertise. J. Exp. Psychol. Gen. 115, 107–117. 11. Pellicano, E., and Rhodes, G. (2003). Holistic processing of faces in preschool children and adults. Psychol. Sci. 14, 618–622. 12. Kanwisher, N., McDermott, J., and Chun, M.M. (1997). The fusiform face area: A module in human extrastriate cortex specialized for face perception. J. Neurosci. 17, 4302–4311. 13. Kanwisher, N. (2000). Domain specificity in face perception. Nat. Neurosci. 3, 759–763. 14. Golby, A.J., Gabrieli, J.D.E., Chiao, J.Y., and Eberhart, J. (2001). Differential responses in the fusiform region to same-race and other-race faces. Nat. Neurosci. 4, 845–850. 15. Rhodes, G., Jeffery, L., Watson, T.L., Clifford, C.W.G., and Nakayama, K. (2003). Fitting the mind to the world: Face adaptation and attractiveness aftereffects. Psychol. Sci. 14, 558–566. 16. Kontsevich, L.L., and Tyler, C.W. (1999). Bayesian adaptive estimation of psychometric slope and threshold. Vision Res. 39, 2729–2737. 17. Watson, T.L., and Clifford, C.W.G. (2003). Pulling faces: An investigation of the face-distortion aftereffect. Perception 32, 1109–1116. 18. Webster, M.A., and MacLin, O.H. (1999). Figural aftereffects in the perception of faces. Psychon. Bull. Rev. 6, 647–653. 19. Haxby, J.V., Hoffman, E.A., and Gobbini, M. (2002). Human neural systems for face recognition and social communication. Biol. Psychiatry 51, 59–67. 20. Rossion, B., and Gauthier, I. (2002). How does the brain process upright and inverted faces? Behav. Cogn. Neurosci. Rev. 1, 63–75. 21. George, N., Dolan, R.J., Fink, G.R., Baylis, G.C., Russell, C., and Driver, J. (1999). Contrast polarity and face recognition in the human fusiform gyrus. Nat. Neurosci. 2, 574–580. 22. Gauthier, I., Tarr, M.J., Moylan, J., Skudlarski, P., Gore, J.C., and Anderson, A.W. (2000). The fusiform “face area” is part of a network that processes faces at the individual level. J. Cogn. Neurosci. 12, 495–504. 23. Rossion, B., Dricot, L., Devolder, A., Bodart, J.-M., Crommelinck, M., de Gelder, B., and Zoontjes, R. (2000). Hemispheric asymmetries for whole-based and part-based face processing in the human fusiform gyrus. J. Cogn. Neurosci. 12, 793–802. 24. Kourtzi, Z., and Grill-Spector, K. (2004). fMRI adaptation: a tool for studying visual representations in the primate brain. In Fitting the Mind to the World: Adaptation and Aftereffects in High-
25.
26. 27.
28.
Level Vision., G. Rhodes and C. Clifford, eds. (Oxford: Oxford University Press), in press. Clifford, C.W.G., and Rhodes, G. eds. (2004). Fitting the Mind to the World: Adaptation and Aftereffects in High-Level Vision (Oxford: Oxford University Press), in press. Hurlbert, A. (2001). Trading faces. Nat. Neurosci. 4, 3–5. Leopold, D.A., O’Toole, A.J., Vetter, T., and Blanz, V. (2001). Prototype-referenced shape encoding revealed by high-level aftereffects. Nat. Neurosci. 4, 89–94. Webster, M.A., Kaping, D., Mizokami, Y., and Duhamel, P. (2004). Adaptation to natural face categories. Nature 428, 557–561.