[Proposal] Are faces smooth surfaces we specialize in?

In Propasognosia—or face blindness—patients not only are unable to recognize faces, but also amoeba-like or potato-like shapes1. To understand how visual perception works or rather what types of features are encoded and represented in the brain, we must ask: what do amoebas or even potatoes have in common with faces? They are all curvy, smooth surfaces . Are faces just a type of smooth surface we specialize in?

Background and rationale

Evidence suggests that faces are processed differently than other objects we are familiar with. Specific mechanisms that are unique to face recognition may exist, as is suggested by face-inversion effects2, composite effects3 and whole-part effects4, that together points toward a holistic framework for face recognition processing5. However, even faces have common properties or features that are shared with other objects. The fusiform face area—a visual area which is highly selective to faces—is also tuned to features such as curvature and contrast6, and natural image statistics greatly vary across dimensions such as curviness and sharpness. Faces are also 3-Dimensional in nature and are characterized by a smooth surface.

Object recognition research has been historically limited to studying 2D processing of objects, namely by strictly using 2D objects and features as stimuli. In addition, most accepted object recognition models rely on the hierarchical combinations of simple 2D features—e.g., edges—that allow invariant object recognition. 2Dness can, however, be considered a projection of some 3Dimensional view of an object and recent electrophysiological evidence in the macaque shows that the visual system is instead representing 3D features of objects7–9.

If our brains are encoding 3Dness of objects, object surface may be a cortical organizational principle for higher-level visual areas—such as category, size and animacy10,11. If this is the case, within a higher-level visual area or across visual areas—e.g., lateral occipital complex, inferotemporal cortices—, the 3D surface of objects would be mapped according to smoothness descriptors. Using psychophysics, we can test if faces are a type of smooth surface we specialize. If the same neuronal population that processes faces is responsible for processing smoothness or smooth surfaces, we should be able to quantitatively measure face recognition thresholds, interference effects and adaptation effects, as detailed in the methods section below. Together, the proposed experiments below aim to test whether smooth surfaces are a meaningful 3D feature of object recognition and whether faces are a type of smooth surface we specialize in.


Experiment 1

In the first experiment, participants will learn the identity of 6 different potato-like smooth shapes. After the learning period, the learned shapes and distractors will be presented in an identification task where participants will be asked to identify the shapes, which will have different sizes, positions and rotations when compared to the learned shapes. In sequence, participants will learn the identity of 6 faces and will be asked to identify them in the same procedure as above. In a separate session, participants will learn the identity of 6 spiky or pointy objects (control condition), that will also be followed by a learning period of the identity of 6 new faces. The order of the two sessions will be randomized across participants. The analysis will include measuring the learning and discrimination rates for faces after either the smooth-shape or spiky-shape learning procedure. Those rates will be measured according to the theory of signal detection (for more information, see12)We predict that people who are good potato-shape discriminators will also be good face discriminators, which would be uncorrelated with how good they are in discriminating spiky-shape identities.

Experiment 2

In the second experiment, we will measure participants’ thresholds for smoothness. In each trial, two similar potato-like shapes will be presented side-by-side with varying smoothness levels. Participants will choose which shape is the smoother of the two. Spiky shapes will again be used as controls—i.e., measuring a threshold for spikiness sensitivity. Our prediction is that people with lower smooth-shape thresholds will be better at recognizing faces and no association with spiky-shape thresholds. We are specifically untapping peoples’ abilities to perceive smoothness and associating it with face identification performance.

Experiment 3

Assuming that the same neuronal population is tapped for both smooth-shapes and faces, we predict that face recognition performance is altered after adaptation to smooth shapes. Therefore, the third experiment will be an adaptation study, in which we expect to observe adaptation effects for face recognition after adapting to smooth shapes, but not spiky shapes. Participants will learn the identity of 6 faces. Next, they will perform the adaptation task, in which they will adapt to several smooth shapes and then be asked to identify a face. Here both reaction times and hit/false alarm rates12 will be analysed. To control for low-level features (e.g., amount of curvature lines), adaptation to textures containing the same image statistics will serve as a first control (e.g., similar amount of curvature but no enclosed surface)13. A second control will entail using spiky-shapes as adaptors. Adaptor order will be randomly displayed within the experiment. We predict that smooth shapes should impair face identification performance, whereas the control conditions should not.

Sample size and budget

Psychophysical face adaptation studies with similar effect sizes typically recruit 10-20 participants14,15 while two-alternative forced choice tasks usually recruit in the order of 100 participants16. Considering the lack of a controlled environment in online testing, we will recruit 200 participants for the first two experiments and 50 for the third, totalling 450 participants. In total we are asking for £5,333.33. This was calculated by estimating each experiment will roughly last up to 1 hour and we will pay £8 per participation. The cost breakdown is as follows: £3,600.00 for participant payments, £1,200.00 for the service fee and an extra £533.33 (including service fees) for bonus payments and recruiting up to 50 participants in case we need to exclude participants—e.g., due to technical difficulties, problems with compliance or internal consistency.

Preregistration and open research

The study has been preregistered using AsPredicted.org and the link for the preregistration is: https://aspredicted.org/kg4tm.pdf. The study’s findings, data, task, and analyses codes will be made available using the Open Science Framework (OSF).


  1. Laeng, B. & Caviness, V. S. Prosopagnosia as a deficit in encoding curved surface. J. Cogn. Neurosci. 13 , 556–576 (2001).

  2. Yin, R. K. Looking at upide-down faces. J. Exp. Psychol. 81 , 141–145 (1969).

  3. Young, A. W., Hellawell, D. & Hay, D. C. Configurational information in face perception . Perception 16 , (1987).

  4. Tanaka, J. W. & Farah, M. J. Parts and Wholes in Face Recognition. Q. J. Exp. Psychol. Sect. A 46 , 225–245 (1993).

  5. Logothetis, N. K. Object recognition: Holistic representations in the monkey brain. Spat. Vis. 13 , 165–178 (2000).

  6. Caldara, R. et al. The fusiform face area is tuned for curvilinear patterns with more high-contrasted elements in the upper part. Neuroimage 31 , 313–319 (2006).

  7. Yamane, Y., Carlson, E. T., Bowman, K. C., Wang, Z. & Connor, E. A neural code for three-dimensional object shape in macaque inferotemporal cortex. Nat. Neurosci. 11 , 1352–1360 (2008).

  8. Vaziri, S., Carlson, E. T., Wang, Z. & Connor, C. E. A channel for 3D environmental shape in anterior inferotemporal cortex. Neuron 84 , 55–62 (2014).

  9. Hinkle, D. A. & Connor, C. E. Three-dimensional orientation tuning in macaque area V4. Nat. Neurosci. 5 , 665–670 (2002).

  10. Konkle, T. & Caramazza, A. Tripartite organization of the ventral stream by animacy and object size. J. Neurosci. 33 , 10235–10242 (2013).

  11. Bao, P., She, L., Mcgill, M. & Tsao, D. Y. A map of object space in primate inferotemporal cortex. Nature (2020). doi:10.1038/s41586-020-2350-5

  12. Heeger, D. Signal Detection Theory. (1997).

  13. Freeman, J. & Simoncelli, E. P. Metamers of the ventral stream. Nat. Neurosci. 14 , 1195–1204 (2011).

  14. Dehmoobadsharifabadi, A. & Farivar, R. Are face representations depth cue invariant? J. Vis. 16 , 6 (2016).

  15. Leopold, D. A., O’Toole, A. J., Vetter, T. & Blanz, V. Prototype-referenced shape encoding revealed by high-level aftereffects. Nat. Neurosci. 4 , 89–94 (2001).

  16. Heathcote, A., Freeman, E., Etherington, J., Tonkin, J. & Bora, B. A dissociation between similarity effects in episodic face recognition. Psychon. Bull. Rev. 16 , 824–831 (2009).