Head pose invariant face recognition using a bank of neural networks

摘要

A method for recognizing faces in relatively unconstrained environments, such as offices, is described. This system can recognize faces irrespective of their orientation and distance relative to the camera. As the pattern recognition mechanism we use a hank of relatively small neural networks of the multilayer perceptron type. Each one of these perceptrons has the task of recognizing only a single person's face. The perceptrons are trained with a set of nine face images representing the nine main facial orientations of the person to be identified, and a set face images from various other persons. In order to treat these images in a unified manner, the center of the neck is extracted as the reference point. Geometric normalization and reference point determination makes use of 3-D data obtained from a stereo camera. The system achieves a recognition rate of about 95 percent.