Photometric stereo system recognizes nose shapes

UK scientists are developing a technique for scanning noses could be a quicker and easier way to verify a person's identity. Unlike other facial features used for biometrics, such as eyes or ears, noses are difficult to conceal and are not changed much by facial expression.

Adrian Evans and Adrian Moorhouse at the University of Bath Dept of Electronic & Electrical Engineering (Bath, UK) and Melvyn Smith and Gary Atkinson from the University of the West of England (UWE; Bristol, UK) Machine Vision Laboratory decided to investigate whether images of people's noses could be used to recognize individuals. They used a photographic system called PhotoFace, developed at UWE, to scan the 3-D shape of volunteers' noses and used software to analyse them according to six main nose shapes: Roman, Greek, Nubian, Hawk, Snub, and Turn-up.

Instead of using the whole shape of the nose, the researchers used three characteristics in their analysis: the ridge profile, the nose tip, and the nasion or section between the eyes at the top of the nose. While the researchers used a relatively small sample, they found that nose scanning showed good potential for use as a biometric, with a good recognition rate and a faster rate of image processing than with conventional biometric techniques such as whole face recognition.

Evans said, “Noses are prominent facial features, and yet their use as a biometric has been largely unexplored. We wanted to find out how good they could be at recognizing individuals from a database.” There's no one magic biometric--irises are a powerful biometric but can be difficult to capture accurately and can easily be obscured by eyelids or glasses."

Noses, however, are much easier to photograph and are harder to conceal, so a system that recognizes noses would work better with an uncooperative subject or for covert surveillance.

Melvyn Smith who led the team at UWE that developed the PhotoFace system, said, “This collaborative project with Bath is very exciting work with great potential. PhotoFace is 3-D face data capture system developed as part of an EPSRC funded project involving UWE, Imperial College, the Home Office (Scientific Development Branch), and General Dynamics."

He added, "It works by taking photos lit by a flash from several different angles so that four images are taken in very rapid succession of every point on the face, each under different controlled lighting conditions. This is known as photometric stereo and UWE's Machine Vision Laboratory is one of only three UK centres with expertise in this area. The software then works out the colour, surface orientation, and depth of each point on the face by analysing the shading within each of the photos."

Above: These figures show an example of a raw image set recovered using the device.

Above: Application of Lambertian photometric stereo then gives the following field of surface normals

Images courtesy: Gary Atkinson, Machine Vision Laboratory, UWE Bristol

Smith notes that the technique is able to achieve a level of detail that is beyond current competing technologies and can be extended to a myriad of other applications, ranging from industrial surface inspection to cosmetics. The researchers plan in the future to build up a larger database of noses to test and refine the software to see if it can pick out individuals from a larger group of people, or distinguish between relatives from the same family.