seems easy - when we open our eyes, we see
large portions of cerebral cortex are devoted to perception
20% cerebral cortex devoted to processing vision
hard to get computers to do what we find easy
Recognition → involves both bottom-up and top-down processing
Bottom-up → information from the sensory receptors (edges, colours, tones), assembled into a whole. Data-driven.
Top-down → information from higher levels of brain (knowledge, expectations.) Concept driven.
Retina → LGN → primary visual cortex → areas beyond primary visual cortex
Early visual system sensitive to basic features (spots light) but as move up the visual system increasing sophistication (orientation, motion, colour, form, faces)
The perception of a whole object is built up from smaller parts of vision
Marr → computational model
Treisman → Feature Intergration Theory
Biederman → Recognition-by-components
Marr: Computational Model
Three stages:
Primal Sketch = Local features stage
2D Sketch = Shapes stage
3D Model representation = Objects stage
Triesman: Feature Integration Theory
2 stages
Preattentive stage: stimulus decomposed into basic properties (primitives.) Automatic, rapid, unconscious, parallel (independent of number distractors), bottom-up Visual search: by measuring time taken to find targets can see which primitives “pop-out”
Intergration stage - once primitives are extracted they are combined (conjunction.) Not automatic, slow, conscious, effort, serial (depends number of distractors, top-down
Biederman: Recognition-by-components
Theory how recognise 3D objects based on building complex objects from set primitives. 36 basic geons - the ‘alphabet’ of objects
Each geon has unique set features that allow it to be easily identified from different points of view
Destroying geons by obscuring their distinguishing features makes it hard to recognise objects
Direct Perception - Ecological
J.J. Gibson (mid 20th century) - rejected need for internal representation
External world (optic flow fields) so rich that perception is ‘direct and immediate’
Subsequent work with computers showed not true - vision is hard.
Ambiguities of Retinal Projections:
Retina image could be due to any of the lines A to E
Top-down processing influences how interpret bottom-up information
Experience creates perceptual sets - readiness to perceive a stimulus in certain way
Constructivist
Sensory systems construct internal representation of world (Helmholtz, 1856)
internal representation mediates perception (Necker cube) - top-down. strong role of past experiences/knowledge
Heavily influenced modern computational theories which stress computation upon representation (A.I.) - Richard Gregrory (1923-2010)
The perceptual system needs work out what stimuli go together to form objects
Basic processes in perceptual organisation
figure-ground
grouping (gestalt laws)
Given unconnected elements (unrelated arrows and Ys) perceptual system forms 2 possible objects even creating subjective contours
Figure-ground
Complex stimulus automatically broken down by perceptual system into figure (emphasized foreground) and ground (background)
figure - has meaning, stands in front, has edges that seperate it from background
Certain properties stimulus influence what area seen as figure and what seen as background
Area size – stimuli with smaller area more likely to be seen as figure
Symmetry – stimuli possessing symmetry more likely to be seen as figure
Meaning - stimuli posessing meaning moree likely to be seen as figure
Grouping
Gestalt psychologists came up with a set of ‘laws’ that predict what tends to group with what tends to be percieved as figure or ground
Gestalt Laws:
Proximity
Objects that are closer to me tend to be grouped together
Similarity
Objects that are more similar to one and other tend to be grouped together. Similarity can be in terms of any diension: shape, size, colour or luminance
Continuity
Sensations that appear to create a continuous form are percieved as belonging together
Closure
People tend to fill in missing contours to form complete objects
Common fate
Objects that move in same direction and speed are perceived as a group.
Camouflage - situations where difficult separate figure from ground
Woodland Encounter by Bev Doolittle (1985)
In recognising the perceptual world we use both bottom-up and top-down processing
Basic processes in perceptual organisation were covered – how components of scene are grouped together (Figure-Ground and Grouping using Gestalt Laws)