Facial image identification using Photomodeler

Abstract

We present the results of a preliminary study on the use of 3-D software (Photomodeler) for identification purposes. Perpetrators may be photographed or filmed by surveillance systems. The police may wish to have these images compared to photographs of suspects. The surveillance imagery will often consist of many images of the same person taken from different angles. We wanted to see if it was possible to combine such a suite of images in useful 3-D renderings of facial proportions.Fifteen male adults were photographed from four different angles. Based on these photographs, a 3-D wireframe model was produced by Photomodeler. The wireframe models were then rotated to full lateral and frontal views, and compared to like sets of photographs of the subjects. In blind trials, 9/15 of the wireframe models were assigned to the correct sets of photographs. In five/15 cases, the wireframe models were assigned to several sets, including the correct set. Only in one case was a wireframe model not assigned to a correct set of photographs at all.