1
Manan P. Shah 2* , Xiaokun Xu 1 , Sarah B. Herald 2 , Irving Biederman 1,2 1Department of Psychology, 2Neuroscience Program, University of Southern California *[email protected] http://geon.usc.edu/ Part-Whole Effect A neurocomputational account of the face configural effect Gabor-Jet Models Conclusions Illustration of overlap of medium- sized receptor fields of two Gabor “jets” with five scales and eight orientations. A jet models aspects of the tuning of a V1 hypercolumn. A difference in a single part appears more distinct in the context of a face than it does by itself. Experiment 3: Can the same theory account for the Face Composite Effect? Yes Experiment 2: Is it large RFs or low SFs? Large RFs Face Composite Effect Experiment 1: Is the configural effect largely produced by cells with large RFs (low SFs)? Yes Identical top halves of two faces look different when their different bottom halves are aligned rather than offset. The face configural and composite effects can be derived from models composed of overlapping receptive fields (RFs) characteristic of early cortical simple-cell tuning but also present in face- selective areas. Because of the overlap in RFs, variation in a single face part or half is propagated to the activation values of large RFs throughout the face. References Face parts (a) and composite target faces (b) created from these parts for the current replication of the part-whole identification experiment of Tanaka and Farah (1993). (c) Predicting response accuracy from large RFs (low SF) versus small RFs (high SF) components in the Gabor-jet representation (via model 1) of faces. A greater proportion of the variance is predictable from the large RF (low SF) components. (a) (b) (c) The configural effect (isolated part vs. composite) as a function of spatial frequency (all SF, high, and low pass). There is only a minimal effect of SF. Therefore the configural effect is produced by large RFs, not by low SFs. Spatial filtering of the part and whole face stimuli in Experiment 2. RF held constant, SF varied. Version 1. Illustration of the computation of dissimilarity for a corresponding pair of jets positioned at nodes in a rectangular grid for a pair of face images. Version 2. Fiducial point version of the Gabor-jet model. Particular jets automatically center themselves on landmark features of a face like the pupil of the right eye. This effectan influence of differences in the lower halves of the facescan be produced by the fiducial point model. Because of a reduction in the overlap of the RFs (perhaps also requiring the context of a face template) from the shift, the influence of the lower half is reduced when it is no longer aligned with the upper half. Above: Dissimilarity computed via Gabor-jet model version 2. Hypothesis The representation of faces (but not objects) retains aspects of the initial multiscale, multiorientation tuning of early cortical visual stages and the configural effect is produced by the overlap of large receptive fields in which a change in the shape of one face part will affect the activation of many cells with large RFs not centered on that face part. Tanaka and Farah 1993 Xu et al., 2014 Xu et al., 2014 Xu et al., 2014 Xu et al., 201 Xu et al., 2012 Xu et al., 2014 Xu et al., 2014 (Nishimura, 2008) (Biederman 2014) (Biederman, 2014) Biederman, I., Xu, X., & Shah, M. (2014). An Account of the Face Configural Effect. Journal of Vision, 14(10), 204-204. De Valois R. L., De Valois K. K. (1988). Spatial vision. Oxford, UK: Oxford University Press. Farah M.J. (1995). Is face recognition “special”? Evidence from neuropsychology. Behavioural Brain Research, 76, 181189. Lades J. C. V., Buhmann J., Lange J., Malsburg C., Wurtz R., Konen W. (1993).Distortion invariant object recognition in the dynamic link architecture. IEEE Transactions on Computers: Institution of Electrical and Electronics Engineers,42, 300311.Nishimura, M., Rutherford, M. D., & Maurer, D. (2008). Converging evidence of configural processing of faces in high-functioning adults with autism spectrum disorders. Visual Cognition, 16(7), 859-891. Nederhouser M., Yue X., Mangini M. C., Biederman I. (2007). The deleterious effect of contrast reversal on recognition is unique to faces, not objects. Vision Research, 47, 21342142. Rossion, B. (2013). The composite face illusion: A whole window into our understanding of holistic face perception. Visual Cognition, 21(2), 139-253. Tanaka J. W., Farah M. J. (1993). Parts and wholes in face recognition. Quarterly Journal of Experimental Psychology A, 46, 225245. Xu, X., Biederman, I., & Shah, M. S. (2014). A neurocomputational account of the face configural effect. Journal of Vision, 14, 1-9. Yue X., Tjan B., Biederman I. (2006). What makes faces special? Vision Research, 46, 38023811.

A neurocomputational account of the face configural effectgeon.usc.edu/~biederman/presentations/ShahBiederman_OPAM2014.pdf · Part-Whole Effect A neurocomputational account of the

Embed Size (px)

Citation preview

Page 1: A neurocomputational account of the face configural effectgeon.usc.edu/~biederman/presentations/ShahBiederman_OPAM2014.pdf · Part-Whole Effect A neurocomputational account of the

Manan P. Shah2*, Xiaokun Xu1, Sarah B. Herald2, Irving Biederman1,2

1Department of Psychology, 2Neuroscience Program, University of Southern California

*[email protected]

http://geon.usc.edu/

Part-Whole Effect

A neurocomputational account of the face configural effect

Gabor-Jet Models

Conclusions

Illustration of overlap of medium-

sized receptor fields of two Gabor

“jets” with five scales and eight

orientations. A jet models aspects of

the tuning of a V1 hypercolumn.

A difference in a

single part

appears more

distinct in the

context of a face

than it does by

itself.

Experiment 3: Can the same theory

account for the Face Composite Effect?

Yes

Experiment 2: Is it large RFs or low SFs?

Large RFsFace Composite EffectExperiment 1: Is the configural effect largely produced by cells

with large RFs (low SFs)? Yes

Identical top halves of two faces look different

when their different bottom halves are aligned

rather than offset.

The face configural and composite effects can be derived from models composed of overlapping

receptive fields (RFs) characteristic of early cortical simple-cell tuning but also present in face-

selective areas.

Because of the overlap in RFs, variation in a single face part or half is propagated to the

activation values of large RFs throughout the face.

References

Face parts (a) and

composite target

faces (b) created

from these parts

for the current

replication of the

part-whole

identification

experiment of

Tanaka and Farah

(1993).

(c) Predicting response accuracy

from large RFs (low SF) versus

small RFs (high SF) components in

the Gabor-jet representation (via

model 1) of faces. A greater

proportion of the variance is

predictable from the large RF (low

SF) components.

(a)

(b)

(c)

The configural effect (isolated part vs.

composite) as a function of spatial frequency

(all SF, high, and low pass). There is only a

minimal effect of SF. Therefore the configural

effect is produced by large RFs, not by low SFs.

Spatial filtering of the part and whole face stimuli in

Experiment 2. RF held constant, SF varied.

Version 1. Illustration of the computation of

dissimilarity for a corresponding pair of jets positioned

at nodes in a rectangular grid for a pair of face images.

Version 2. Fiducial point version of

the Gabor-jet model. Particular jets

automatically center themselves on

landmark features of a face like the

pupil of the right eye.

This effect—an influence of differences in the

lower halves of the faces—can be produced by

the fiducial point model. Because of a reduction

in the overlap of the RFs (perhaps also

requiring the context of a face template) from

the shift, the influence of the lower half is

reduced when it is no longer aligned with the

upper half. Above: Dissimilarity computed via

Gabor-jet model version 2.

Hypothesis

The representation of faces (but not objects)

retains aspects of the initial multiscale,

multiorientation tuning of early cortical visual

stages and the configural effect is produced

by the overlap of large receptive fields in

which a change in the shape of one face part

will affect the activation of many cells with

large RFs not centered on that face part.

Tanaka and Farah 1993

Xu et al., 2014

Xu et al., 2014

Xu et al., 2014

Xu et al., 201

Xu et al., 2012

Xu et al., 2014

Xu et al., 2014

(Nishimura, 2008)

(Biederman 2014)

(Biederman, 2014)

Biederman, I., Xu, X., & Shah, M. (2014). An Account of the Face Configural Effect. Journal of Vision, 14(10), 204-204.

De Valois R. L., De Valois K. K. (1988). Spatial vision. Oxford, UK: Oxford University Press.

Farah M.J. (1995). Is face recognition “special”? Evidence from neuropsychology. Behavioural Brain Research, 76, 181–189.

Lades J. C. V., Buhmann J., Lange J., Malsburg C., Wurtz R., Konen W. (1993).Distortion invariant object recognition in the dynamic link architecture. IEEE Transactions on

Computers: Institution of Electrical and Electronics Engineers,42, 300–311.Nishimura, M., Rutherford, M. D., & Maurer, D. (2008). Converging evidence of configural processing of

faces in high-functioning adults with autism spectrum disorders. Visual Cognition, 16(7), 859-891.

Nederhouser M., Yue X., Mangini M. C., Biederman I. (2007). The deleterious effect of contrast reversal on recognition is unique to faces, not objects. Vision Research, 47, 2134–

2142.

Rossion, B. (2013). The composite face illusion: A whole window into our understanding of holistic face perception. Visual Cognition, 21(2), 139-253.

Tanaka J. W., Farah M. J. (1993). Parts and wholes in face recognition. Quarterly Journal of Experimental Psychology A, 46, 225–245.

Xu, X., Biederman, I., & Shah, M. S. (2014). A neurocomputational account of the face configural effect. Journal of Vision, 14, 1-9.

Yue X., Tjan B., Biederman I. (2006). What makes faces special? Vision Research, 46, 3802–3811.