Face Recognition

1
Face Recognition

• The following slide has photographs of people.
• Try to identify the people.
• You may not recognise all of them.
• If you cant identify them, do you perhaps know
  anything about one of them?
• Write down your answers. Do it completely on
  your own to get the best benefit from the exercise

2
(No Transcript)
3
Face Recognition

• Now, choose any one of the faces, whether it is
  one you know or one that you are unfamiliar
  with, and write a description of that face.
• Again, do this on your own.

4
Here are the pictures again ask your neighbour
how well you did.
5
Face Recognition

• Turn to page 320 in Pennington. Read the study
  by Bahrick, and briefly state
• Aim Method Results Conclusion

6
Bahrick - continued

• Difference between recognition and identification
  over time also investigated.
• Table page 320.

7
Application of Bahrick

• Important application of face identification
  research is in crime detection e.g. photofits.
  
• Photofits are based on forward facing, stationary
  images.
• Bahrick showed that identification is best when
  actual person is there. Cant always do that, so
  what could be an alternative to photofit that
  would be more efficient?

8
Ripper Images
9
Ripper Images
10
Photo identification

• Bahrick showed that identification is best when
  actual person is there. Cant always do that, so
  what could be an alternative to photofit that
  would be more efficient?
• Modern methods use video technique. Instead of
  flat pictures, composite video images are
  prepared with the witness. Seems to be an
  improved success rate.

11
Feature Analysis

• P 321 Pennington
• Look up description of photo you made earlier
• Shepherd, Davies Ellis (1981) free recall
  recognition test.
• Mainly features described/recalled

12
Feature recall
Hair
Nose
Eyes
Mouth
Chin
eyebrows
Forehead
13
Shepherd Davis Ellis showed that features are
usually recognised in the order given in
the Previous slide.
Descriptions of faces fit in more with
externalfeatures, but when we know the person,
we useinternal features.
How do your descriptions fit in with this
feature detection theory?
Feature detection is a theory based on
recognitionof individual features. The
combination of the featuresproduces a whole
image with meaning. Question Topdown or
bottomup theory?
Gollum
14
Holistic-form Theory P322 Pennington.

• Generally recognised that facial recognition is
  more complicated than simple feature analysis
• See someone you know in the street, need to refer
  to previously stored knowledge about the person
• Top-down or bottom-up?

15

• Holistic-form theory suggests that face is
  recognised as a whole (holistic
• based on the whole) by analysis of
• relationship between features
• feelings aroused by the face of each person we
  know
• semantic information about the person (e.g.a
  name?)

Ellis suggests we have a facial template for each
person we know. We see them and match data with
the template. This is Top-Down
16
Holistic-form processing Young and Hay (1986)
17
Holistic-form processing Young and Hay (1986)
18
Holistic-form processing Young and Hay (1986)

• Renee Zellweger and Liv Tyler
• But Composites are not so easy to identify
• Feature analysis would be just as easy to
  identify composites if only features are used

19
Holistic-form processing Young and Hay (1986)

• But Holistic-form would suggest it is harder to
  identify composites because we are in effect
  making a new face out of the parts.
• Similar effects with cut up and upside down bits
  of faces
• Be Amazed!!

20
(No Transcript)
21
(No Transcript)
22
That was George Bush as weve never seen him! It
is known as the Thatcher Illusion guess why
23
(No Transcript)
24
The moral of the story is

• If you want to rob a bank, get a wig!

25
Application to Investigation of Crime

• Photofit based on building up face from
  individual features.
• Operators intensively trained.
• When they tried to train potential witnesses
• Feature training resulted in poorer performance
• Suggestion is
• Faces are stored holistically rather than by
  features

26
Familiar Face
27
Familiar Face
Structurally encoded A mental description Or
representation of The face is produced From the
stimulus
28
Familiar Face
Structurally encoded A mental description Or
representation of The face is produced From the
stimulus
Activates Face Recognition Unit (FRU) Each face
known to the viewer has FRU containing Structural
information About the face.
29
Familiar Face
Structurally encoded A mental description Or
representation of The face is produced From the
stimulus
Activates Face Recognition Unit (FRU) Each face
known to the viewer has FRU containing Structural
information About the face.
Activates Person Identity Node
(PIN) Information about the person, e.g.
occupation, usual context, Do I like him/her, etc.
30
Familiar Face
Structurally encoded A mental description Or
representation of The face is produced From the
stimulus
Activates Face Recognition Unit (FRU) Each face
known to the viewer has FRU containing Structural
information About the face.
Activates Person Identity Node
(PIN) Information about the person, e.g.
occupation, usual context, Do I like him/her, etc.
Activates Name Generation Name is stored
separately, Accessed last. (TOT State?)
31
Young, Hay and Ellis (1985)

• Pennington page 325
• Aim Method Results Conclusion

32
Young, Hay and Ellis (1985)

• Aim to test holistic model
• Method diary study
• Results (i) no reports of naming without prior
  information. (ii) some cases, (19) occupation
  but not name. (iii) more cases, (23)
  familiarity, but no more.
• Conclusion supports holistic model sequence.

33
Recognition Disorders

• Prosopagnosia very rare cant recognise
  familiar faces. Own reflection?
• Sufferers get vague emotional feeling of
  recognition (I ought to know them!) but no
  conscious awareness of knowledge.

34
Recognition Disorders

• Capgras syndrome
• Doubles have replaced people I know
• Recognition, but emotionally, it is not the
  person.
• Tested by GSR emotion response.
• No difference between friends and strangers

35
Recognition Disorders

• What conclusion from disorder studies?
• Face recognition depends on more than just face
  patterns.
• So, Holistic model is more likely.
• Implication in e.g. identity parades.