Effecting an improvement to the fitness function. How to evolve a more identifiable face Charlie Frowd* 1 , Joanne Park 2 , Alex McIntyre 2 , Vicki Bruce 3 , Melanie Pitchford 1 , Steve Fields 4 , Mary Kenirons 1 and Peter J.B. Hancock 2 1 Department of Psychology, University of Central Lancashire, Preston, UK 2 Department of Psychology, University of Stirling, Stirling, UK 3 College of Humanities and Social Science, University of Edinburgh, Edinburgh, UK 4 Department of Psychology, University of Leicester, Leicester, UK *Corresponding author: Dr Charlie Frowd, Department of Psychology, University of Central Lancashire, Preston PR1 2HE. Email: cfrowd@uclan.ac.uk. Phone: (01772) 893439. Abstract Constructing the face of a criminal from the selection of individual facial parts is a hard task. We have been working on a new system called EvoFIT that involves the selection and breeding of complete faces. The approach is theoretically better founded and produces more identifiable composites than those from a traditional ‘feature’ system. In the current paper, we explored three new methods of presenting faces to a person using EvoFIT. A better quality face was evolved if (1) the external parts of the face were subjected to a Gaussian blur, allowing a user to focus on the important inner region of the face, (2) if the faces were simplified to make them sketch-like in appearance, and (3) if users took longer in deciding which faces to select for breeding. Taken together, these approaches would appear to make a marked improvement in the ability to evolve an identifiable likeness of a target. 1. Introduction Witnesses to and victims of crime are often required to build a visual likeness of a person’s face. Such an image is known as a facial composite and is traditionally achieved by the selection of individual facial features: eyes, nose, mouth, etc. There is considerable research to suggest that this procedure does not produce identifiable images [1-6], the root cause of which is the procedure used: we are not good at constructing a face in such a piecemeal fashion [7]. We have been working on a new method for witnesses to produce the face of a criminal [2-6, 8-13]. It is a software program called EvoFIT and was conceived [2] to more closely reflect the manner in which faces are perceived, as wholes [14-16]. In brief, witnesses are presented with sets of faces, 18 per screen, and select a few that look most like the target. The selected faces are combined together to produce a new generation of faces. By repeating the selection and breeding process a few times, a good likeness of a target can be evolved. In laboratory tests, composites from EvoFIT are more identifiable than composites from a traditional ‘feature’ system [5,6,10,12]. There are other similar systems to EvoFIT – ID (Cape Town, SA) [26] and EigenFIT/EIFT-V (Kent, UK) [27] – although their effectiveness appears to be unknown. At the heart of EvoFIT is a pair of face models, one each for facial shape (feature shape and position on the face) and facial texture (greyscale information). Each pair is built from Principal Components Analysis (PCA) from photographs of whole faces, and combined to synthesise a face for a given race, age and gender. The EigenShape and EigenTexture coefficients, which themselves represent a face in PCA space, are randomised at the start but combined during breeding with uniform crossover and mutation operators. Early work on EvoFIT was concerned with the best method to present faces to witnesses and on fine tuning the evolutionary parameters; a detailed account is presented in Frowd et al. [3] and at BLISS 2007 [13]. In brief, this led to users being presented with facial shape, facial texture, and then opting for the best shape/texture combination: a superior face with the highest perceived fitness in that generation. While the selected faces are subjected to crossover and mutation operators, and with equal selection pressure, the superior chosen face is given twice the number of breeding opportunities and is carried forward to the next generation as part of an elitist approach. The shape/texture combination, and the subsequent increase in selection pressure and elitism, produced a basic