This research paper focuses on EvoFIT, a facial composite system developed by Frowd et al. (2004) and now used in police investigations to reconstruct facial images of alleged perpetrators. The system employs Principal Components Analysis (PCA), as PCA can create a large number of faces from very few components. EvoFIT currently uses a database consisting of 72 faces and this study aimed to determine whether increasing or decreasing the variability within the system’s face-space could result in a more accurate facial composite.
Participants were asked to construct faces from memory in either a small (36 faces), medium (72 faces), or large (144) database, and to either set the height and width ratio of the faces to be displayed, or not set this ratio. Composites were subsequently identified using ‘spontaneous’ and ‘constrained’ naming tasks. Results indicated that composites constructed using the small (36) database resulted in more ‘spontaneous’ correct naming than the medium database, but no differences were apparent with the large database, or with any of the databases using the constrained naming task. Possible explanations and implications for further research are discussed.
Keywords: Facial Composite, Principal Components Analysis, EvoFIT, Face-space.
- There are currently no refbacks.
Copyright © 2014 University of Central Lancashire