This technique can address the correspondence problem by generating from the lighting source artificial features, including even non-featured regions such as the cheek, chin and forehead. However, because the active sensing technique uses a visual beam projector to project a structured pattern onto the face, the person whose face image is to be acquired may feel a strong aversion such as to the beam glare.In this paper, we present a developed a nonintrusive 3D face modeling system for random-profile-based 3D face recognition. This system consists of a stereo vision system and a rotatable near-infrared line laser (NILL) [17]. Because the proposed system uses NILL, which is not visible, the person whose face image is to be acquired perceives nothing of the 3D acquisition.

The system can not only reconstruct full-range full-density 3D face data, but it can also directly reconstruct precise face profiles. Using both full-density 3D face data and 3D face profiles, we propose random-profile-based 3D face recognition that is pose-invariant. In this context, the term random means that the number of profiles (from 7 to 12) and the extracted profiles from the face region are arbitrary. Moreover, before addressing the face recognition technique in this paper, we introduce a face feature-based registration process, which is much faster than conventional iterative closest points (ICP) for making two sets of 3D face data into the same pose. We next explain the face recognition, which is performed by comparing the 3D random face profile as a probe and the 3D full-density 3D face data.

Because 3D random-face profiles are used as the probe, this technique is memory efficient and faster than using full-density 3D face data, while outperforming 2D face recognition under pose variation.The rest of this paper is organized as follows. In the next section, we briefly review related works on 3D acquisition systems and 3D face recognition. In Section 3, we present a proposed system that includes a 3D face data acquisition AV-951 system, registration, and random-profile-based 3D face reconstruction. The details of our experiments and results are presented in Section 4. Finally, our conclusions are presented in Section 5.2.?Related Works2.1. Three-dimensional (3D) Face Acquisition SystemA variety of 3D face acquisition systems have been developed to acquire accurate 3D face data. These systems can basically be divided into three kinds of systems: stereo vision systems (SVS) [10,11], laser scanners (LS) [14,16], and structured light systems (SLS) [12,13]. The stereo vision system, which consists of two optical sensors [10], reconstructs 3D data by performing a triangulation between the corresponding points of images captured from each camera [15].