Zusammenfassung der Ressource
Recognition
- Visual recognition
- 1.Description (construct internal representaion)
- 2.processes to store information
- 3.processes to compare object
viewed with stored descriptions
- 4.different angles (nature of mechanism)
- generating & comparing descriptions of current
object viewed with previous object descriptions
seen.
- Humphreys & Bruce (1989)
- 5-step model
- PERCEPTUAL PROCESSING (what it is)
- SEMANTIC CLASSIFICATION (object specifics)
- NAMING (what object is)
- VIEWPOINT-DEPENDANT OBJECT DESCRIPTIONS
(similar to Marr's 2 1/2D sketch)
- EARLY VISUAL PROCESSING
(similar to Marr's full primal sketch)
- humans CLASSIFY between items
- 'object' (an apple)
- class of object
- specific (eg, Sigmund Freud)
- specific instance
- BETWEEN CATEGORY v's WITHIN CATEGORY
- faces researched differently to object recognition
- Pike et al. (2002)
- can identify famous faces even if
poorly presented (eg E-FIT)
- recognition of unfamiliar faces is poor
(eg, line-ups)
- Young et al. (1993)
- different cognitive functions used
to recognise faces compared to
emotions displayed
- Gibson (1986)
- Active processing (interact with
environment)
- recognition through touch
- feedback system = brain
+ touch receptors)
- where pressure applied
is regulated by brain.
Based on sensory
information generated
by touch
- location of limbs
(KINESTHESIS) using touch
receptors can guide hands
- exact position
(kinesthesis) + what it
feels like (touch
receptors) = HAPTIC
INFORMATION
(generate description of
object)
- haptic info good for recognition of object's weight. Yet visual
more accurate for processing shape (esp. 3-D) & distance (out of
reach)
- haptic is active perception. Not always
passive, therefore, criticism of H & B model
- keep track of relative
location on limbs
(PROPRIOCEPTION)
- Lederman & Klatzky (1987)
- exploratory procedure (use of hands to
gain haptic information.
- (1990) each procedure gained particular
type of info (shape, texture) by moving
hands over contour/surface
- recognising 2-D processing may use
different cognitive processing than
3-D
- Marr & Nishihara (1978) recognising 3-D objects
- generalised cones (3-D images)
generating an object-centred description.
canonical coordinate frame.
- multi-step process
- 1. derive the object shape.
(identify central axis using 2 1/2
D sketch)
- has 3 assumptions, must accept all
3 to correctly identify.
- 2. locate object axis and derive 3-D description
- 1.area of concavity. 2.divide area in
primitives. 3.find axis for each primitive.
4.link them to form 3-D
- 3. compare 3-D description to mental catalogue of objects
- 1. compare mental catalogue.
2. hierarchical levels. 3. once
match found process ends
- evaluation
- FOR
- locating central axis critical to
recognition- supported by Lawson &
Humphreys (1996)
- Warrington & Taylor (1978)
patients with right hemisphere
focal lesions- difficulty
recognising from different
viewpoints. eg. comparing
photos of same object from
different viewpoint.
- unable to convert 2-D to 3-D. Features
needed to identify was obscured by different
angle (rotation)
- Humphreys & Riddoch (1984) foreshortened
images/ hidden features. foreshortened
recognised less- suggesting major axis is
important to forming 3-D model
- explains misinterpretation if contour
generator is misidentified
- AGAINST
- within-category discrimination
hard to explain- converting to
generalised cones leads to same
representations (cannot distinguish
differences)
- Pattern matching theories ( uses
templates in memory) unlikely
though as could have too many
templates or large generic
template
- Feature recognition
theories (key
features of image
extracted)
- sturctural description
theories ( key features &
how they are organised
with each other)
compared to internal
representations until
match found
- adapts to variety & ambiguity.
describe in computer & human
language. can recognise 3-D
versions of 2-D images
- 3-D recognition must be able to
occur independent of viewpoint.
viewer-centred description
change to object-centred
description
- Beiderman's Theory
- agrees with Marr & Nishihara's theory
- Enter text here
- complex objects represented
as hierarchies of similar shapes
- approx 36 geons used to
represent objects
- concavity used to sub-divide objects