Toshiba develops system that quickly extracts faces from video

by Dhiram Shah

Toshiba Corp. has developed a technology for detecting human faces quickly and with high precision from video. The technology involves an algorithm that operates on a standard personal computer to analyze the images captured with a small camera. The system uses just a 300,000-pixel resolution camera, but if the room is properly lighted the algorithm can confirm the human face from a distance of 2-3 meters, even when the person is slowly walking in the room. The face can be detected and the person tracked. The algorithm can differentiate human faces from background with a precision of 97 per cent, compared to the standard 92 per cent, and it can perform this task 160 per cent faster than conventional algorithms. Faces are recognized by searching a number of parameters related to the shading due to facial features including the eye, cheeks, nose and nostrils. The company hopes to have a practical version of the technology
ready for such applications in 12 months.