Dr Stuart Gibson, Dr Christopher Solomon andDr Matthew Maylin.
Identification from Facial CompositesIdentification from Facial Composites
(A New Approach)(A New Approach)Nuffield Foundation, Eyewitness Identification EvidenceNuffield Foundation, Eyewitness Identification Evidence
Contact: [email protected]
Affiliation: School of Physical Sciences, University of Kent.
Feature Based Approach to Facial Composite Construction Witness describes suspect’s face to operator
(cognitive interview). Appropriate facial features are selected from
databases contained within the facial composite system.
Features positioned to achieve a likeness to the suspect.
Image refined – better likeness
Operator Training: Cognitive Interview Fisher and Geiselman 1992. 1. Recreating the context: mental imagery,
sounds, smells, feelings relating to the event. 2. Focused concentration: Persuading the witness
to concentrate on the task of remembering the face.
3. Multiple retrieval attempts: Multiple attempts can unlock previously un-recovered detail.
4. Varied retrieval: e.g. Try a different chronological order.
Motivation for New ApproachMotivation for New Approach Witness finds it difficult to recall descriptions
individual faces – weak point in the composite procedure.
Configuration of features important – feature based system achieves this in an ad-hoc manner. (Turner et al 1999)
Seek a technique based on recognition of whole faces.
EFIT-V: New Methodology for EFIT-V: New Methodology for Composite ConstructionComposite Construction EFIT-V system components:◦ Appearance model (Cootes et al)◦ Evolutionary algorithm
Whole face approach Does not rely heavily on the witness’ ability to
verbalise descriptions of features On average composites are generated
relatively quickly Improved image quality Easy to use
Commercial contact:Commercial contact:VisionMetric Ltd VisionMetric Ltd enquiries@[email protected]
EFIT-V: Outline MethodEFIT-V: Outline Method
GenotypeVector of appearance
model parameters
PhenotypeGenerated facial image
MutationChanging of appearance
Model parameters
Deployment (user) Feedback
Instances of deployment ◦ “…witness or victim believes that they would
recognise an offender again, even if they cannot recall facial features in detail”. V. Burgin, Principal Forensic Services Officer, Derbyshire Police Constabulary
Facial ID Officers recovered images from witnesses 50% of the time compared to only 25% when using the feature based system.
ReferencesReferences◦ Young et al, “Configurational information in face perception”. Perception, 16:747-
749, 1987.◦ Caldwall et al, “Tracking a criminal suspect through face-space with a genetic
algorithm”. Proc. Of the Fourth International Conference on Genetic Algorithms, 416-421, 1991.
◦ Gibson et al, “The generation of facial composite systems using an evolutionary algorithm”. Recent Advances in Soft Computing, July 2006.
◦ Gibson et al, “Synthesis of photographic quality facial composites using evolutionary algorithms”. Proc. Of the British machine Vision Conference 2003, 1:221-230, 2003.
◦ Hancock, “Evolving faces from principal components”. Behaviour Research Methods, Instruments and Computers, 32(2):327-333, 2000.
◦ Pallarea-Bejarano et al, “Eigenfit: Building photographic quality facial composites using evolutionary algorithms”. Proc. Of the Third IASTED International Conference on Visualization, Imaging and Image Processing, p1, Sept 2003.
◦ T. F. Cootes and C. J. Taylor. Modelling object appearance using the grey-level surface. In E. Hancock, editor, 5th British Machine Vison Conference, pages 479–488, York, England, Sept. 1994. BMVA Press..
◦ Turner et al, “Making faces: Comparing e-fit construction techniques”. Proc. Of the British Psychological Society, 7(1):78, 1999.