Signal and Image Processing for Biometrics

Chapter 53D Face Recognition ¹

 

 

5.1. Introduction

Three-dimensional (3D) face recognition allows us to deal with some problems related to the pose and lighting conditions. In fact, the 3D information, once obtained through appropriate sensors, is invariant to changes in lighting and pose conditions. Nevertheless, the facial deformation caused by expressions has been one of the challenges that researchers and manufacturers are trying to address. In addition, 3D face recognition requires the 3D acquisition of faces. Not only commercial 3D sensors, but also the solutions proposed by the research community have limitations. These include the range of the sensors, that is 1–2 m, the controlled lighting conditions, the precision, and ultimately the duration of the acquisition.

There are currently two major paradigms of face recognition using the 3D modality: the symmetric recognition where the data in the gallery and the probe data are similar, specifically 3D or 3D + texture, and the asymmetric recognition that uses heterogeneous data from the gallery and from the probe. Thus, the gallery consists of 3D or textured 3D data while the probe data are only texture images or vice versa. The advantage of the latter paradigm is that the use of 3D information is limited. It is also referred to as recognition assisted by 3D.

The structure of this chapter follows to a certain extent the order of steps of the 3D recognition, from the acquisition to the recognition. First, we present the current ...