Lecture 2 Notes: Memory and Cognition: Perception

Perception
​ What is it?
​ Why is perceiving an object so difficult?
​ What cues does the visual system use to perceive objects
​ Connecting anatomy and philology to perception
​ Separating figures from the ground
Sensation vs. Perception – What’s the difference?
​ Detection of energy vs. Interpretation of sensory information
​ Bottom-up vs. top-down processing
​ Perception involves
​ 1) Current sensory input
​ 2) Previous Experience X 2
Low-Level Vision-Outline
​ What's your vision?
​ Problem to be solved
​ How simple computations yield more complex information
Problems to be solved
​ Many of the qualities of objects that we would like to know about trade-off with
other qualities
​ Shape/Orientation
​ Reflected/Light Source/Shadow
​ Size/Distance
Size/Distance
​ Monocular Depth Cues
​ Relative size
​ Texture gradient
​ Interposition
​ Linear perspective
​ Height in plane
​ Binocular Depth Cues
​ Binocular disparity
​ Binocular convergence
Problems
​ Problem 1 is the inverse projection
​ Problem 1 is that the types of information that we want to trade off with one
another
​ Problem 2 is that the initial information the visual system has is extremely
impoverished
How do you get from one to another
​ Vision is not an exact representation of what is in the world, it is a representation
of what is probably in the world
Why Edges
​ An edge is a sudden discontinuity in intensity
 ​ Edges frequently correspond to the boundaries of objects; a map of edges is a
good start to identifying objects
​ Edges are invariant to lighting conditions
​ How to find
​ Computationally easy to find discontinuities
​ Compare means of adjacent columns, rows, diagonals
Biological Evidence
​ 120 million rods 6 million cones → 1.2 million ganglion cells
​ Ganglion cells: center during
​ One cell and off-cells
​ On cells when light is focused on that receptor field they become active +
when light is focused in the center they inhibit activity in the area around it
+ see inhibition
Visual Detection: Shape and Contour
​ Detecting lines and edges
​ Simple cells-orientation-specific slits of light in a particular location
​ It does seem that some of the cells relatively early in the visual processing
stream care about the edges
Real World problems to which we can apply Gestalt Principles
​ Basic tendency to organize visual input
​ Segmentations
​ Determining where objects are in an image and what their boundaries are
​ Grouping
​ Grouping stuff together as part of the same objects; for example, across
occluders
​ Gestalt approach
​ Problems-we don't perceive local events in an image and perceive
more global figures
​ Elucidate principles that determine the grouping of local “things” in
an image into figures
​ Proximity
​ Similarity
​ Collinearity
Figure-ground Organization
​ Perceive a boundary belonging to a “thing-like” featural region
​ It has a definite shape, and appears closer, whereas the ground appears
farther and extends behind the figure
​ Size, Symmetry
​ Frame of Reference
​ Ex. align a rod within this frame so that the rod is vertical
Object Recognition
​ Human perception is more than the sum of information in the distal stimulus
​ Perceiving figures is critical for object recognition
​ What is object recognition-identify a complex arrangement of sensory stimuli, and
view it as distinct from its background
Models of Object Recognition
​ Template
​ Feature
​ “New wave” of feature models (3D features)
Template Model
​ 1) Memory representation is a holistic unanalyzed entity (a template).
​ 2) An input pattern is compared to the stored representation.
​ 3) Identity is determined by the selection of the template with the greatest
amount of overlap.
​ Problems:
​ Size
​ Orientation
​ Need too many templates
Feature-Analysis Models
​ 1) Inputs are broken down into a small list of distinctive features
​ 2) Identity is determined by selecting the feature list most similar to the input
​ Gibson (1969). Subjects were given a reaction test to determine if two letters were
the same or different.
​ G vs W
​ P vs R
​ RT = 458 msec
​ RT = 571 msec
​ Good:
​ Larsen & Bundenson (1996) – developed a model that identified 95% of
addresses and zip codes
​ Visual input does seem to be decomposed into features
​ Physiological evidence about edges (Hubel & Wiesel 1965)
​ Problems:
​ No role for the spatial relation between features
​ Moving objects appear differently
​ Natural objects
New Wave of Feature Models
​ Recognition by component models uses three-dimensional features.
​ 36 Geons lead to 30,000 readily discriminable objects.
​ You usually only need to see the edges of a geon
​ Geons have properties that are invariant to rotations, size, and translation
Problems with Geons
​ Mental rotation plays a role
​ Local view hypothesis
Top Down vs. Bottom Up
​ Top-down processing is strong when a stimulus is registered for just a fraction of
a second.
​ Top-down processing is also strong when stimuli are incomplete or ambiguous.
​ Object recognition combines bottom-up and top-down processing
Word Superiority Effect
​ Faster and more reliable in identifying a letter when it is part of a word than a
non-word or in isolation
Face Perception
 ​ Should be a challenging task
​ Need to recognize faces from different angles, in different settings, with different
expressions
Alternatives
​ Nature of the task
​ Perceptual expertise
​ Specific to faces (Haxby and colleagues?)