Chapter 11: Perceiving Objects and Scenes

How do we organize visual scenes?

• Designing a “perceiving machine”

• Vision is complex - early “self-driving” vehicles used cameras

• Current “self-driving” vehicles used cameras

• No vehicle is truly “self-driving” - they have advanced driver assisted systems (NTSB)

• We’re going to focus on human vision

What are problems identifying objects?

• Stimulus on the receptors is ambiguous

• Objects can be hidden or blurred

• Objects look different from different viewpoints

Inverse Projection Problem

• Image on retina is 2D

• Objects are 3D

• Many objects can create same image on retina

• How do we know which is the correct image?

• Ambiguous

Hand Shadow Puppets

• Good example of vision with limited info

• Create image from flat shadow

Facial Recognition

• Viewpoint Invariance - recognize face from different viewpoints

• Humans are very good at this

• Ability peaks in the 30s for facial recognition

• Difficulty with different racial groups than your own

Computer Facial Recognition

• Can recognize you even if face is blurred

• Looks at multiple faces, clothing, etc.

• Performs worse than non-blur

Theories of Object Recognition

• Gestalt principles of organization

  • “The whole is different from the sum of the parts”

  • Emphasis on top-down

• Recognition by components theory

  • Geons - basic components of objects

  • Emphasis on bottom-up

Origins of Gestalt

• Apparent motion

  • Max Wertheimer - perception arises from sensations

  • No actual motion involved, so how can the perception of motion arise?

Gestalt Principles

• Principles of Perceptual Organization

  • Pragnanz - good figure, simplest figure is best

    

  • Closure - small gaps are overlooked

  

  • Similarity - similar items get grouped

  

  • Good Continuation - follow a smooth curve rather than abrupt

  

  • Proximity - items close together get grouped

  

  • Common Region - items seen as in same region get grouped

  

  • Uniform Connectedness - similar properties get grouped

  

  • Synchrony - items that occur at same time get grouped together

  

  • Common Fate - items moving in same direction get grouped

• • Meaningfulness/Familiarity - things that form familiar patterns get grouped

Perceptual Segregation

• Also known as figure ground segregation

• What is the figure and what is the ground

• Must look at properties of figured and grounds

• Useful to look at reversible figure ground images

Figure Ground

• What makes the figure?

• What makes the ground?

• Shared contours

• Border Ownership - whoever owns the border is the figure

Reversible Figures

• Symmetry

• Size

• Orientation

• Meaning

  

Gestalt Summary

• Heuristics not principle or algorithms - rule of thumb

• They’re quick

• They’re correct most of the time but not all

• Reflect properties of the environment - they work

• Used frequently to make logos and artwork

Recognition by Components Theory

• Geons - “geometric ions”

• Basic components of objects

• Proposed by Irving Biederman

• 36 different geons - represent most shapes

• Accidental versus non-accidental viewpoints of objects

Object Viewing

• Accidental Viewpoint - a viewpoint an object is not normally viewed at

• Non-accidental property - a property that occurs except when you straight on or at accidental viewpoint e.g. curved edge of coin

• Ex: 3D sidewalk drawings use accidental viewpoint

Non-Accidental Properties (NAP)

• Each geon has a unique set of NAPs

• Discriminability - each geon can be discriminated from others

• Principle of Componential Recovery - We can identify an object if we identify its geons