Hancock PJB, McIntyre AH & Kittler J (2009) Caricaturing to improve face matching. In: Bio-inspired Learning and Intelligent Systems for Security (BLISS), 2009 Symposium on Print. Bio-inspired Learning and Intelligent Systems for Security (BLISS), 2009 Symposium on Print, Edinburgh, 20.08.2009-22.08.2009. Los Alamitos, CA: IEEE Computer Society Press, pp. 49-49. http://www.proceedings.com/07116.html; https://doi.org/10.1109/BLISS.2009.17
Abstract Identity verification by matching face images is a common security task; is this person on a wanted list? With unfamiliar faces, this is surprisingly difficult, with error rates in the region of 30%. With photographic identification increasingly common, it is important to find ways that may reduce the chances of mistakes. Facial caricatures by cartoonists grossly exaggerate the most characteristic features of a familiar face yet tend to be easily identifiable. Here we investigated whether exaggeration of unfamiliar faces through the application of caricature would facilitate discrimination and enhance identification in face matching tasks. Images were caricatured by morphing them 30%, 50% and 70% away from an average face image, thereby exaggerating the ways in which each face image deviated from a facial norm. We found a significant reduction in false matches, without affecting hit rates. The method is being incorporated in a computerised tool to assist with face matching.