|
Face recognition is a key function of the human brain...let me put it another way, from a very, very early age we can all recognise faces, from the familiar view from mother's knee to spotting a friend in a crowd. It arises as a main modal of biometrics because of increased global security concerns and needs. While the technology is fast become highly effective, it is not perfect. Computers too can process a digital image and compare it with a database entry to carry out simple face recognition - but only if the light is right. Throw in a few shadows, sunlight through a window, or a flickering overhead fluorescent light, and the computer usually cannot spot the difference between John Doe and Joe Bloggs. The Problem is most obvious in face-based biometric authentication and has been a focus of research for three decades. Now, researchers in China have turned to near infra-red to help computers cope with variable lighting conditions and so recognise even the most shadowy of faces. Stan Li of the Institute of Automation at the Chinese Academy of Sciences, and his students RuFeng Chu, ShengCai Liao, Lun Zhang has devised a novel approach that renders variations in ambient illumination irrelevant to the face recognition system by exploiting an active near infrared (NIR) imaging system. By using NIR, the team was able to "encode intrinsic information in a face" and for the first time derived illumination invariant representation of faces using advances they made in both hardware device and software algorithms. While the hardware produces face images subject to a degree of freedom in monotonic transform to a greyscale image,the application of local binary pattern (LBP) features compensate to produce a representation of a face that will be almost exactly the same regardless of illumination. The team has used a statistical learning algorithm to extract the most distinguished features from a large pool of invariant LBP features and construct a highly accurate face-matching engine. Their system has achieved impressive accuracy and speed. In a practical setting, the team has demonstrated just how robust their system is. It can even cope well with the specular reflections from active NIR lights on spectacles, as well as the person's movements and different ethnic groups. The technology will enable much more accurate and quicker face recognition for so-called cooperative user applications, in which the person is willingly showing their face to allow them to be scanned and compared with their travel documents, for logging into a computer, or withdrawing funds from an automated teller machine, ATM, at the bank. "The system has been deployed, since June 2005, at Shen Zhen-Hong Kong border control, the largest border-crossing point in the world where on average 400 thousands crossings happen everyday. About 150 gates were equipped with this NIR face recognition technology for Hong Kong residents to use as fast, self-service immigration checking channels. They have been more preferred than the traditional manned checking by most users," Li told us.
Related links:
|
|
