The view tolerance of human identity recognition depends on horizontal face information
- Psychological Sciences Research Institute (IPSY), UCLouvain, Belgium
- Université Clermont-Auvergne, CNRS, LAPSCO, France
- Statistical Methodology and Computing Service (SMCS), UCLouvain, Belgium
- Institute of Neurosciences (IoNS), UCLouvain, Belgium
Figures
Figure 1
A graphic illustration of the horizontal structure that emerges from the average of multiple views of a face.
The tolerance of human face identity recognition to drastic appearance variations caused by varying lighting, viewpoint, facial expression, etc. has been proposed to emerge through averaging (Burton et al., 2005). With increased exposure to a face, an averaging mechanism would progressively whiten accidental variations in appearance while preserving stable cues to identity. Past illustrations (Mike Burton, 2013; Burton et al., 2005) used varying lighting and expressions but moderate pose variations. Here, we show that averaging highly diverse views (from left to right profile) of a face produces a horizontally smeared image, which suggests that, across encounters with a face, cues at orientations other than horizontal are whitened. Images of two celebrities (George Clooney and Daniel Radcliffe) were sampled from the internet. Image averages were made of 40% left-averted, 40% right-averted, and 20% frontal views, in line with exposure to face views in natural viewing (Oruc et al., 2019). The luminance and RMS contrast of the averaged faces were set to a luminance of 0.5 and contrast of 0.4. Using this procedure, one can appreciate the emergence of the so-called bar code, namely, the vertical arrangement of horizontally oriented cues which carries the natural statistics of the face category and of face individual identity (Dakin and Watt, 2009).
Figure 2
Stimulus conditions.
Columns. Each identity was viewed from seven different viewpoints ranging from +75° to –75° in steps of 25°. Rows. All images were filtered in the Fourier domain to preserve only a selective range of orientation, from 0° (vertical) to 157.5° in steps of 22.5°.
Figure 3
Human face identity recognition: orientation tuning across viewpoint.
(A) Sensitivity of human observers (n = 22) to facial identity (d’) as a function of the orientation filter (0° to 180° in 22.5° steps), and face viewpoint (yaw: +75° to –75° in 25° steps). Dots and error bars represent mean d’ values and 95% confidence intervals across participants. Solid lines and shaded areas indicate the mean posterior predictions and 95% credible intervals from the Gaussian Bayesian multilevel model. (B) Population-level mean parameters of the Gaussian Bayesian Multilevel model: peak location, standard deviation, base amplitude, and peak amplitude. Each estimate is plotted with 95% credible intervals as a function of face viewpoint. The 95% credible intervals reflect the uncertainty of the model. They indicate a 95% probability that the true population parameter lies within the specified range, given the observed data.
Figure 4 with 1 supplement
Sensitivity (i.e. performance in the recognition task) of human (n = 22) and model observers (n = 22 each) and image energy across viewpoints and orientations.
Left column. 3D surf plots of the normalized energy/sensitivity across orientation and viewpoints. Middle column. Matrix representations of the normalized energy/sensitivity across orientation and viewpoints. Right column. Matrix representations of the Pearson correlation (non-Fisher Z-transformed) of the normalized orientation distributions of energy/sensitivity across viewpoints.
Figure 4—figure supplement 1
Distribution of sensitivity d’ of human, view-selective model, view-tolerant model, and view-average model observers, all viewpoint and orientation conditions confounded.
Figure 5
Fisher Z-transformed Pearson partial correlation of the orientation sensitivity profiles between humans and each model (n = 22), while controlling for the alternate model and image energy profiles.
Error bars show the 95% confidence intervals. The faded gray line depicts the maximally achievable correlation for the separate viewpoint conditions in the human dataset (see Methods for details).
Figure 6
View-average model observer.
(A) Image averages of the different views of each face identity of the stimulus set. (B) Recognition performance (normalized sensitivity d’) of the view-average model observer across views and orientations. From left to right: 3D surf plots of the normalized sensitivity across orientation and viewpoints, matrix representations of the normalized sensitivity across orientation and viewpoints, and matrix representations of the Pearson correlation of the normalized orientation distributions of sensitivity across viewpoints.
Tables
Table 1
Posterior mean and 95% credible interval for each parameter of the Gaussian model, at each viewpoint.
| Parameter | Viewpoint | Estimate (posterior mean) | 95% Credible interval – lower bound | 95% Credible interval – upper bound |
|---|---|---|---|---|
| Peak location | –75 (left profile) | 99.5 | 90.02 | 108.98 |
| –50 | 94.4 | 86.98 | 101.87 | |
| –25 | 90.62 | 83.58 | 97.73 | |
| 0 (full front view) | 90.99 | 88.22 | 93.78 | |
| 25 | 91.98 | 85.00 | 99.01 | |
| 50 | 88.19 | 80.82 | 95.51 | |
| 75 (right profile) | 80.24 | 70.73 | 89.76 | |
| Standard Deviation | –75 (left profile) | 40.37 | 27.40 | 54.4 |
| –50 | 46.51 | 34.51 | 59.29 | |
| –25 | 47.2 | 35.75 | 59.07 | |
| 0 (full front view) | 44.94 | 40.55 | 49.41 | |
| 25 | 51.36 | 40.04 | 63.14 | |
| 50 | 50.02 | 38.18 | 62.23 | |
| 75 (right profile) | 45.06 | 32.5 | 58.12 | |
| Base amplitude | –75 (left profile) | 0.81 | 0.29 | 1.31 |
| –50 | 0.44 | –0.16 | 0.99 | |
| –25 | 0.11 | –0.49 | 0.66 | |
| 0 (full front view) | 0.07 | –0.2 | 0.34 | |
| 25 | –0.09 | –0.72 | 0.5 | |
| 50 | 0.27 | –0.34 | 0.84 | |
| 75 (right profile) | 0.65 | 0.11 | 1.16 | |
| Peak amplitude | –75 (left profile) | 1.01 | 0.47 | 1.58 |
| –50 | 1.68 | 1.10 | 2.30 | |
| –25 | 1.87 | 1.29 | 2.48 | |
| 0 (full front view) | 1.72 | 1.49 | 1.97 | |
| 25 | 2.07 | 1.45 | 2.72 | |
| 50 | 1.81 | 1.22 | 2.45 | |
| 75 (right profile) | 1.12 | 0.56 | 1.71 |
Table 2
Mean and 95% confidence interval of the (Fisher Z-transformed) partial correlation coefficients between human and each model orientation d’ profiles while controlling for the variance in image energy and alternate model.
| Model | Viewpoint-specific | Viewpoint-tolerant | ||
|---|---|---|---|---|
| Mean | 95% CI | Mean | 95% CI | |
| +075 | 0.06 | [–0.09 0.2] | 0.49 | [0.30 0.67] |
| +050 | 0.04 | [–0.10 0.18] | 0.5 | [0.34 0.66] |
| +025 | 0.56 | [0.42 0.71] | 0.3 | [0.17 0.42] |
| 000 | 0.79 | [0.62 0.97] | 0.19 | [0.09 0.30] |
| –025 | 0.36 | [0.18 0.54] | 0.34 | [0.17 0.51] |
| –050 | 0.1 | [–0.06 0.25] | 0.59 | [0.45 0.72] |
| –075 | –0.05 | [–0.17 0.08] | 0.56 | [0.35 0.77] |
Table 3
Difference of the human-model partial correlation coefficients between viewpoint-selective and view-tolerant model observers.
Positive t values indicate a stronger correlation with the viewpoint-selective model, and negative t values indicate a stronger correlation with the view-tolerant model.
| 95% CI for mean difference | |||||
|---|---|---|---|---|---|
| View | Mean difference | Lower | Upper | t | p Holm |
| +075 | –0.47 | –0.86 | –0.07 | –4.20 | 0.002 |
| +050 | –0.43 | –0.82 | –0.03 | –3.82 | 0.01 |
| +025 | 0.23 | –0.16 | 0.63 | 2.08 | 1 |
| +000 | 0.69 | 0.30 | 1.09 | 6.21 | <.001 |
| –025 | 0.03 | –0.37 | 0.42 | 0.23 | 1 |
| –050 | –0.61 | –1.00 | –0.21 | –5.44 | <.001 |
| –075 | –0.78 | –1.18 | –0.39 | –7.02 | <.001 |
Additional files
-
Supplementary file 1
Prior distributions of the four parameters of the Gaussian Bayesian multilevel model.
- https://cdn.elifesciences.org/articles/108495/elife-108495-supp1-v1.docx
-
MDAR checklist
- https://cdn.elifesciences.org/articles/108495/elife-108495-mdarchecklist1-v1.docx
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
The view tolerance of human identity recognition depends on horizontal face information
eLife 14:RP108495.
https://doi.org/10.7554/eLife.108495.4
