Public Funding for Aurora Computer Services Limited

The aim of the “FIRMA” project is to produce a novel face recognition engine that, uniquely, is capable of matching facial images extracted from “uncontrolled” sources. This would have an immediate impact in security applications such as passport agency checks and police suspect evaluation/identification. “Uncontrolled” means images that have not been captured specifically for facial recognition (FR), such as a street scene or an un-posed photo. In these images the orientation of the face could be at any angle and exposure or lighting equally variable. All current FR technologies require carefully posed images to achieve any dependable accuracy. Lighting conditions must be uniform, the face must be flat to the camera, and expressionless. In 2007 Aurora invented a novel method of image capture using powerful infrared flash. This work eliminated the dependence of the recognition engine on lighting conditions. In 2010 the company successfully produced another novel recognition engine to match this infrared image against a colour image stored in a person’s ePassport. To successfully match two fundamentally dissimilar images, infrared and colour, the company had to further invent a series of novel mathematical techniques. Accurate recognition is currently being achieved down to 40 inter-eye pixels and requires a very small signature, or template file, for each individual, resulting in very fast processing. Aurora believes that similar techniques could also be innovatively applied to the fundamental problem in facial recognition; dealing with different head angles. i.e. faces not flat to the camera. If successful the benefit this project will provide is a step change in image recognition performance impacting on all applications of face recognition. The outcome of FIRMA will increase accuracy of matching variable head angle, “real world” images, by at least an order of magnitude and potentially change the whole facial recognition market landscape.