3 = Perceptual organisation and object categorisation

• Perception of form

  Before deciding what objects are, it is necessary to identify which parts of a scene contain the forms to-be-recognised

  Psychology started as the study of visual elements (Weber, Fechner, 1890s)The Gestalt Psychologists (Wertheimer, Kofka & Kohler, 1912) proposed that ‘the whole is other than the sum of the parts’ i.e. relationship between elements matters

  Proposed laws of perceptual organisation [from gestalt psychology] (we should think of these as heuristics) – descriptive (not predictive), short cuts that the mind applies to information.
• What is Gestalt psychology?


  A thing that is defined by its form not its elements/content i.e. a square is about the form that the lines make, not the lines themselves.
• ‘Laws’ of perceptual organisation

  Praegnanz

  similarity

  proximity

  good continuation

  closure

  common fate

  familiarity

  NB: these ‘laws’ (heuristics) are descriptive, not explanatory
• ‘Laws’ of perceptual organisation - Praegnanz = ‘good figure’


  Of several geometrical possible organisations, the most simple and stable will be perceived
• ‘Laws’ of perceptual organisation - Similarity and proximity


  We also group with similarity (similar things group together) and proximity (near things group together)
• ‘Laws’ of perceptual organisation - Good continuation 


  Points on straight or smooth lines belong together (like praegnanz, but praegnanz refers to objects, whereas good continuation is just lines)

   
• ‘Laws’ of perceptual organisation = Closure

  Often aided by cues of good continuation, we tend to ‘see’ objects as complete, despite occlusion. And we may even ‘see’ occluders that aren’t there, as in the Kanizsa triangle
• ‘Laws’ of perceptual organisation - Common fate


  Things that move together belong together
• ‘Laws’ of perceptual organisation - Meaningfulness/familiarity


  Things are more likely to form groups if those groups are meaningful or familiar
• Key concepts: your brain reflects the world

  Humans evolved (and develop) in certain environments; our brains incorporate regularities of these environments as assumptions or heuristics

  Heuristics (something that under normal, everyday conditions, works most of the time; can assume that something is the way that it is?) of perceptual organisation make our perception more efficient under normal conditions

  ... but can mislead us when normal conditions do not apply (visual illusions)

  In the real world, their benefits far outweigh their costs
• Heuristics of visual perception means that ...


  Heuristics of visual perception refer to the mental shortcuts or rules of thumb that our brains use to interpret and make sense of visual information quickly and efficiently. These heuristics help us to organize and understand the complex visual stimuli we encounter in our environment.
• Figure-ground segregation - phenomenon


  The figure is seen as the ‘thing’ in the scene, standing in front of the ground, with the ground extending uniformly behind it. The figure-ground border ‘belongs’ to the figure.

  If there is a continuous (closed) boundary, the space it contains will tend to be seen as the figure.
• Figure ground segregation phenomenon continued 


  Regions with symmetry tend to be seen as the figure; non-symmetric = background.

  The smaller region tends to be seen as the figure.

   All other things being equal, the lower portion tends to be seen as figure (84% of the time in 30s free viewing).
• Figure ground segregation phenomenon continued


  Meaningful shapes tend to be seen as the figure, so figure-ground segregation does not strictly precede object recognition; recognition interacts with, and influences figure-ground segregation
• Top-down and bottom-up influences


  Our perceptions do not just emerge from the analysis of sensory data (bottom-up), but are actively constructed, using context, knowledge and expectation (top-down)

  This ‘knowledge’ may be flexibly acquired: Or it may be deeply engrained, possibly even hard-wired (‘top-down’ does not imply ‘conscious’) and our ‘perceptual set’ is updated continuously according to the context, and new information
• How does object representation work in the human brain?


  Low-level: Edges, Spatial frequencies, Colour

  Mid-level: Figure/background, Surfaces. In 3D -Gestalt rules and perceptual organisation

  High-level: Categorical knowledge
• Key concepts: perception is prediction 


  Due to speed of processing, what we see if always slightly (100ms) behind what is occurring in reality.

  We make predictions about the state of the world, and continuously refine them by comparing to sensory data

  These predictions (and the prediction error) are what we actually experience
• Top-down knowledge: belief, expectation desire. -> predicitons -> (<- prediction error) comparison <- bottom up (sensory data).

  A) top down

  B) bottom up

  
• Perceptual inference can be thought of as Bayesian


  Bayes work on conditional probabilities describes how we should adjust our beliefs according to new evidence

  Prior probability: (how much do we expect something) (top-down)

  AND

  Sensory data: (how much evidence do we have for it) (bottom-up)
• Key points from this lecture

  In the early stages of form perception, we ‘parse’ the scene into regions, and distinguish figure from ground

  The brain uses inbuilt ‘heuristics’ to guide interpretation – these reflect regularities of the environments we have experienced

  Perception reflects the interplay of bottom-up (data-driven) and top-down (knowledge-driven) processes

  This interplay can be modelled as a (Bayesian) process of rapidly refining our perceptual predictions in light of the sensory evidence