0010 Week 3 — Object Recognition & Face Perception

What are the four core problems of object recognition per Marr & Nishihara (1978)?

Feature extraction, feature generality, object constancy, generalization vs discrimination

What is the feature extraction problem in object recognition?

How is an object description derived from features or parts in the retinal image?

What is the feature generality problem?

Whether the same set of features can describe objects from very different classes (basic features occur across many categories, giving a common starting point)

What is the object constancy problem?

How an object produces a stable representation despite changes in viewpoint, illumination, and occlusion

What is the generalization vs discrimination problem?

How a representation groups different objects into the same category while still distinguishing similar objects

Contrast object-centered and viewpoint-centered theories of object recognition.

Object-centered: representation describes structural properties + part relations that stay stable across viewpoints (focus on the object). Viewpoint-centered: match the current image to stored views (focus on the perceiver); recognition gets harder as the view differs from stored views.

What is a canonical view?

A frequently encountered view of an object that is recognized more efficiently than novel views

What is the full processing pipeline of Biederman's Recognition-by-Components (RBC) model (1987)?

Edge structure (contours, junctions, curvature, parallelism) → Geon identification (~36 simple 3D volumetric parts) → Structural description (which geons present + their spatial arrangement) → Object representation (match to stored object model)

What are geons?

Simple 3D volumetric parts (~36 of them) that combine to form objects; a small set of parts can describe many objects

What are nonaccidental properties, and why do they matter in RBC?

Properties that don't change with viewpoint (e.g., collinearity, symmetry); they let geons be identified across different viewpoints

In RBC, what determines object identity?

Both the geons present AND their spatial arrangement

How does RBC explain recognition from novel viewpoints?

Geons are recovered via nonaccidental properties that are viewpoint-invariant, so the structural description holds across views

How does RBC explain recognition of partially occluded or degraded objects?

As long as enough geons/vertices survive to recover the component structure, recognition succeeds

What did the geon-deletion studies show about vertices?

Recognition suffers most when deletion disrupts vertices — the component structure becomes hard to recover

Name three strengths of the RBC model.

Small part set describes many objects; supports recognition across novel viewpoints; robust to partial occlusion/degradation

Name three limitations of the RBC model.

Geons don't distinguish objects with similar part structures; fine within-category distinctions need metric shape, texture, or color; recognition can still vary with viewpoint; faces/highly similar exemplars need more detailed representations

What did Shepard & Metzler (1971) find with mental rotation?

Response time increased approximately linearly with angular disparity, suggesting an analog transformation between orientations — evidence recognition reflects experience with particular views

What do the Shepard & Metzler (1971) results imply about viewpoint-centered recognition?

Familiar/canonical views are recognized more efficiently; novel orientations carry a cost

Name two unresolved problems for viewpoint-centered theories.

How many views must be stored; how the system interpolates between familiar views; how different views are linked to a common object identity

Why are object-centered and viewpoint-centered accounts considered complementary rather than mutually exclusive?

Structural representations support cross-viewpoint recognition while view-specific representations preserve orientation/appearance information; their relative contribution depends on task, object category, and brain region

What did Vuilleumier et al. (2002) find with repetition suppression in the fusiform gyrus?

Left fusiform: repetition suppression across both same and different viewpoints (viewpoint tolerant). Right fusiform: suppression mainly when the viewpoint repeated (viewpoint sensitive) — evidence object recognition depends on multiple representations, not a single system

What are the three competing explanations for face processing?

1) Faces depend on dedicated (specialized) mechanisms. 2) Faces place strong demands on within-category expertise + configural processing. 3) Both specialization and experience contribute.

What are the four lines of evidence in the face-processing debate?

Behavioral (part-whole + inversion effects), neuroimaging (category-selective ventral visual cortex), prosopagnosia (selective impairment), development (early biases + effects of experience)

What is holistic/configural processing of faces?

Faces are processed as wholes, not as independent parts

Describe the part-whole effect (McKone et al., 2013).

Study "This is Larry," then test a part alone (which nose was Larry's?) vs the same part in a whole face (which face contains Larry's nose?) — memory is better in the whole-face context for faces; the house/door control shows context doesn't help the same way for objects

What is the face inversion effect (Yin, 1969)?

Turning faces upside-down impairs recognition disproportionately more than inverting other objects — inversion disrupts configural processing

What is the Thatcher illusion, and what does it reveal?

Grotesque distortions of eyes/mouth are obvious upright but hard to notice inverted — upright faces are processed configurally, inverted faces piecemeal

What is the fusiform face area (FFA), and who identified it?

Face-selective response in the fusiform gyrus of ventral temporal cortex; Kanwisher et al. (1997)

What did O'Craven & Kanwisher (2000) show about the FFA?

Imagery and attention modulate the FFA response

What did the FFA vs PPA comparison reveal?

A double dissociation of category selectivity in ventral visual cortex: FFA responds preferentially to faces, PPA (parahippocampal place area) to places/scenes — evidence for category-selective processing

What is prosopagnosia?

Difficulty identifying people from their faces despite intact vision and intellect; recognition via voice or other cues may remain; the patient can know who someone is but fail to recognize their face

Why is prosopagnosia important evidence in the face-specialization debate?

A selective face-recognition deficit argues faces rely on specialized mechanisms

What did Farah et al. (1995) find with patient LH?

LH's inversion effect was reduced/absent — little difference between upright and inverted faces, unlike controls (big upright advantage); suggests LH lacks the configural processing that inversion normally disrupts

What did Busigny & Rossion (2010) study?

Further prosopagnosia work on holistic processing

What is the expertise hypothesis of face processing?

Face effects reflect within-category expertise, not face-specific modules

How do Greebles test the expertise hypothesis?

Gauthier et al. (1997, 1998, 1999, 2000): training with novel objects (Greebles) produces face-like effects — inversion sensitivity and FFA recruitment — suggesting experience can create face-like processing for non-faces

How do dog/bird/car experts contribute to the debate (Diamond & Carey, 1986)?

Experts show face-like configural effects for their domain of expertise, supporting the idea that expertise (not faces per se) drives the effects

What is perceptual narrowing (Pascalis et al., 2002)?

Infants initially discriminate faces broadly (e.g., human AND monkey faces); with experience, discrimination narrows to familiar face types. Longer looking at a novel face = discrimination.

What is the other-race effect (Spangler et al., 2013), and how is it related to expertise?

Recognition is more accurate for faces from familiar racial groups; experience shapes the difference; it can be reduced with experience/training — consistent with perceptual expertise

What is the "level of analysis" counterpoint (Greene & Rohan, 2025; Behrmann, Gauthier & Tarr, 1999)?

Maybe it's the subordinate/within-category discrimination level that faces require — not faces per se — that drives face-like effects

Why might "faces are special" and "face processing reflects expertise" not be mutually exclusive?

Evidence summary: part-whole + inversion → upright faces benefit from configuration; face-selective activity → some regions prefer faces; prosopagnosia → face ID selectively disruptable; object expertise → experience can produce face-like effects; infant development → experience shapes discrimination — so both specialization and experience contribute