Constructing a three-dimensional face model by a stereo method and its application to optical parallel face recognition

The authors have proposed and fabricated the FARCO (Fast Face Recognition Optical Correlator) based on Vanderlugt Correlator(1) with a super-higher-speed of 1000 faceS/S-2'3. Using its rapid data processing capability, a robust recognition system can be constructed by registering three-dimensional (3-D) face data. There are a number of techniques to obtain 3-D data. Since the system deals with people, it is desirable that instant measurement and non-contact data processing are secured. To meet these requirements, we took photographs and created database of facial images adopting three digital cameras based on the stereo method. Three simultaneously-photographed images were given angular variations from top to bottom and from left to right by 3-D modeling by use of the software. Using our database and these facial images with angular variations, we proceeded on to examination of the system performance, following the algorithm for recognizing images in motion. The paper begins with the process of constructing, the first database for facial images using 3-D model, with those image files made by three cameras. Photographs were taken from the front and to right and left 15degrees. The 3-D model is constructed by these facial images. Making the database from the 3-D model, two-dimensional (2-D) facial images are clipped from part of the model at shifting in each 5degrees from top to bottom, and then from left to right. As a result, the facial images with angular variations were correctly recognized in the database. Through the assessment, the 3-D modeling system proved to be most effective.