EXAM: Object recognition

Question 1

What are structural description models of object recognition

Answer 1

Structural description theories of object recognition propose that we recognize objects by analyzing their structure, breaking them down into simpler parts and then understanding how those parts relate to each other.These theories focus on creating an internal representation of an object’s shape and spatial arrangement, independent of the viewer’s perspective.Essentially, they suggest we build a mental model of an object’s structure and then match that model to a stored representation in memory.

Question 2

INTRO

Answer 2

• Define object recognition and introduce structural description models.
• State that both Marr (1982) and Biederman (1987) propose that recognition involves analyzing an object’s structure and parts.

Thesis: While both theories share a similar viewpoint-invariant, bottom-up framework, they differ in complexity, volumetric parts used, and empirical support—and both face limitations.

Question 3

PARAGRAPH 1 - MARR’S COMPUTATIONAL MODEL (1982)

Answer 3

Marr proposed a multi-stage, bottom-up process:
1. Primal sketch – detects basic edges and contrast.
2. 2.5D sketch – depth and surface layout, viewpoint-dependent.
3. 3D model representation – full, abstract, viewpoint-invariant model using generalised cylinders.

Influential for integrating neuroscience, computer vision, and psychology.

Question 4

PARAGRAPH 2: BIDERMAN’S RECOGNITION BY COMPONENTS MODEL (1987)

Answer 4

Proposes that object recognition occurs through detecting and assembling 36 geons—simple 3D shapes like cylinders, cones.

Emphasizes non-accidental properties (e.g., curvature, edges) and viewpoint-invariant recognition.

Question 5

PARAGRAPH 3: SIMILARITIES BETWEEN THEORIES

Answer 5

• Both are structural description models with a bottom-up, hierarchical process.
• Both argue recognition is viewpoint-invariant at the final stage.
• Use volumetric primitives: geons (Biederman) vs. generalised cylinders (Marr).

*Propose that objects are recognized by matching perceptual input to stored abstract representations in memory.

Question 6

Key point: how do people use the volumetric parts to recognise objects

Answer 6

By matching perceptual input to stored abstract representations in memory

Question 7

PARAGRAPH 4: DIFFERENCES BETWEEN THEORIES

Answer 7

• Scope: Marr covers early to late visual processing; Biederman focuses specifically on recognition.
• Primitives: Marr uses a generalised cylinder; Biederman uses a limited set of geons.
• Complexity: Marr’s model is computationally intensive, while Biederman’s is simpler and more empirically testable.

Question 8

CRITICISMS

Answer 8

• Both theories:

o Overemphasize bottom-up processing, ignoring top-down influences (e.g., expectations, context).

o Do not fully explain recognition for all object classes (e.g., distinguishing dog breeds or faces).

o Focus on static objects, ignoring motion and learning effects.

Evidence from neuroimaging and behavioral studies (e.g., Gauthier & Tarr, 2016; Zimmermann & Eimer, 2013) shows recognition can be both viewpoint-dependent and viewpoint-invariant, depending on task and familiarity.

• Visual perception is not always accurate—contradicting Marr’s assumption that vision “tells the truth.”

Question 9

CONCLUSION

Answer 9

• Both models provide valuable frameworks for understanding how we recognize objects based on structural parts and spatial relations.
• Marr offers a comprehensive theoretical foundation, while Biederman presents a practical, testable model.
• However, a complete theory of object recognition must also include top-down influences, learning, and contextual flexibility not addressed by either model.

Question 10

Neuroimaging evidence

Answer 10

Evidence from neuroimaging and behavioral studies (e.g., Gauthier & Tarr, 2016; Zimmermann & Eimer, 2013) shows recognition can be both viewpoint-dependent and viewpoint-invariant, depending on task and familiarity.

Question 11

Object recognition definition

Answer 11

The process of identifying objects by providing labels or names based on visual information.

Question 12

What pathway is associated with object recognition? Key regions in it?

Answer 12

The Ventral Pathway

This runs from the Occipital lobe to the temporal lobe
It is called the “what” pathway

Key brain regions in this pathway include:

Lateral Occipital Complex (LOC): Involved in processing the shape and structure of objects.

Fusiform Gyrus (especially the Fusiform Face Area, FFA)