object and face recognition Flashcards
gestalt psychology
- Gestalt means “whole” (in German)
- They suggested that perception could not be done by breaking it down into parts, but by considering the whole experience
- The whole is other than the sum of its parts.
- Obvious links to object identification
gestalt laws
Law of Pragnanz (i.e. Good figure - Law of simplicity)
- The percept you see should be the simplest interpretation of the scene
- They are not really “laws”
- Not like laws of physics or the law of natural selection
object ground segmentation
considered as an inference problem about where a given object is throughout a video sequence.
object recognition
Arguably one of the most difficult tasks the visual system has to perform
Clear dichotomy between the human ability to perform a task effortlessly and its intrinsic complexity
Computer scientists have been trying to solve this problem for decades!
How do we do it?
marr’s vision
provides a framework which is both computationally traceable and biologically plausible in the understanding of human vision.
Object parts are represented independently of their spatial configuration and viewpoint.
image based models
Specific views are ‘stored’ and recognition performance is (somehow) based on generalization from these
Image-based models encode ‘structured’ templates of
viewpoint-dependent representations
structural description models
Information about the 3D structure of an object is extracted from a single view
e.g. you see a teapot as a collection of different parts
recognition by component
Objects are represented using basic shape units called geons.
geons
Geons are the simple 2D or 3D forms such as cylinders, bricks, wedges, etc, corresponding to the simple parts of an object in Biederman’s theory of object recognition.
They are defined by variations in a small number of basic parameters called Non-Accidental Properties (NAP)
NAP
• NAPs are basic features that define variation in shapes which are view independent
Properties of Naps
• Curvilinearity – curviness in the 2D image
• Parallelism – lines in parallel in 2D object
• Cotermination – two or more edges that terminate
at same point
• Symmetry – axis of symmetry in 2D image reflect the axis of symmetry on object
• Collinearity – a straight line in the 2D image
Role of disparity and view generalisation during object recognition
Two main theories prediction:
Image based models
Structural description models
canonical viewpoint
describe perspectives in which identification performance is best.
results revealed a double dissociation between task (recognition and recall) and type of object perspective.
The “front”, “side”, and “top” views of an object.
Frequency hypothesis
Maximal information hypothesis
frequency hypothesis
The easiness of recognition is related to the number of times we have see the objects from each viewpoint.
maximal information hypothesis
Maximal information hypothesis: Some views provide more information than others about the objects.
e.g. Clocks are preferred as purely frontal
Why is face recognition important to psychology
- It involves ‘within-category’ discrimination
- Not is it a face, but which face it is?
- Errors in face-recognition can have catastrophic consequences
- Eye witness testimony (e.g. Devlin, 1976).
- Technology
is face recognition ‘special?’
- (i.e. is it independent of object recognition)
- Johnson and Morton (1991), new-born babies will preferentially view faces from day 1 (9 minutes actually!)
- Expression analysis seems to be innate (Meltzoff and Moore, 1977) - though we already accept that this is independent of recognition.
featural hypothesis
hypothesis suggests that faces are primarily remembered due to their facial features (Garner, 1978).
the configurational (spacing) hypothesis
places the emphasis on the relationship amongst the facial features (Bartlett & Searcy, 1993; Diamond & Carey, 1986).
the holistic hypothesis
takes the face as a perceptual whole where both configurational and featural information are required for accurate recognition (Tanaka & Sengco, 1997).
• The holistic hypothesis still places configuration as more important than the features
• However, it does emphasise that loss of either type of information could be detrimental to efficient face perception
upright faces
Yin (1969) - Participants are better at recognising upright faces than they are other objects.
But They are worse for inverted faces than they are for other inverted objects.
The thatcher illusion:
what does the thatcher illusion tell us
• we perceive faces in terms of the global configuration of facial features
• we are unable to detect – or accurately process – the properties of local individual face parts if upside-down
Effects like the ‘Thatcher’ illusion have been taken as evidence that faces are processed holistically
model of face recognition
Functional face recognition model:
• Faces are structurally encoded
• Encoded information activates face recognition units (FRUs)
• If match between encoding and FRU then semantic information can be accessed
• Personal identity nodes (PINs) contain information about that person
using FMRI
subject view faces for a while, then pictures of objects
One area becomes more active during face-viewing (Fusiform face area); another area becomes more active during object-viewing (Lateral occipital complex)
face recognition is special
There are several pieces of evidence that suggest that face recognition might be special:
• Infants show a tendency to track moving faces, at just 9 minutes old
• Face agnosia (prosopagnosia) without object agnosia; object agnosia without prosopagnosia (more on this next lecture)
• Human fMRI result of objects vs. faces (previous slide)
inversion effect
- Healthy participants are better at recognizing upright rather than inverted faces – This effect is not as strong with objects
- Prosopagnosics participants are better at recognizing inverted rather than upright faces
