Horizontally split facial composites show improved identification

1
Horizontally split facial composites show
improved identification

• Alex H. McIntyre, Peter J.B. Hancock and Charlie
  D. Frowd
• Department of Psychology
• University of Stirling

2
Facial Composites

• Images produced by combining separate facial
  features.
• Created with sophisticated software that are
  capable of generating very good likenesses.
• In spite of this identification rates are very
  poor.
• Important and accurate information with the
  composites is not readily perceived.

3
Accuracy of information

• Facial composites are produced from an often
  fleeting memory.
• Not all of the information they contain will be
  accurate.
• May be the only tangible clue in a police
  investigation.
• How can we capitalise on the more accurate
  elements?

4
Perceiving Faces

• Featural information.
• (eyes, nose, mouth etc.)
• Configural information.
• (the relative size and location of features
  relative to one another)
• Information is processed separately and
  differentially.
• In normal adult face perception configural
  information tends to dominate.

5
Effects of Inaccuracy

• Featural inaccuracies are easier to discern.
• Configural inaccuracies are much more subtle but
  exert profound effects.
• They can actually inhibit recognition of accurate
  facial features.
• An inaccurate configuration can even be perceived
  as a completely new face.

6
Configural inaccuracy in facial composites

• 32 original composites.
• 32 configurally enhanced (morphed) composites.
• 32 participants
• viewed 16 of each, with each target represented
  once.
• Conditional hit rate.

7
An original composite and a morphed composite of
Brad Pitt

• 16.6 32.4
• t(31) 6.45, plt.001.

8
Results

• Configural inaccuracy does have a significant
  effect on the identification of facial
  composites.
• How can we reduce the effects of configural
  inaccuracy in an applied setting?

9
Young, Hellawell, Hay (1987)
10
Getting easier?
11
Young, Hellawell Hay (1987)
12
Continued.
13
Young, Hellawell, Hay (1987)
14
Reducing the effects of inaccuracy

• 32 original composites.
• 32 composites horizontally split below the eyes
  and misaligned.
• Direction of split counterbalanced.
• 32 new participants
• viewed 16 of each with each target represented
  once.
• Conditional hit rate

15
Split composites

• Original 17.8
• Split 24.1
• F(1,30) 5.39, plt.05
• Disruption of inaccurate configurations?
• Multiple processing of parts?

16
Type or quantity of processing?

• 32 original composites.
• 32 split-morphed composites.
• 32 new participants
• viewed 16 of each with each target represented
  once.
• Conditional hit rate.

17
Split-morph composites

• Original 15.4
• Split-morph 24.6
• F(1,30) 20.87, plt.001
• (Morph 32.4)
• Disruption of an accurate configuration inhibits
  identification.

18
Composite naming
19
Conclusion

• Splitting facial composites produces a
  significant gain in identification rates.
• Requires just a simple image editor to make a
  significant impact on the successful detection of
  crime.
• The practical implication is clear, police forces
  should consider releasing split composites to the
  public, in order to boost the likelihood of an
  identification.

20
Comparison across experimentsMixed factor ANOVA

• Main effect of composite type (original or
  treated)
• F(1,93) 55.4, plt.001.
• No effect of Experiment, pgt.05.
• Significant interaction, F(2,93) 4.00, plt.05.
  
• All treated composites were identified
  significantly better than the originals, plt.05.
• Identification of original composites did not
  differ across experiments (pgt.05)
• Naming of split-morphs (24.6) split composites
  (24.1) were comparable, pgt.05, but significantly
  poorer than of morphed composites (32.4)
• t(31) 2.70, plt.05.

21
Student project within subject design.