Face Detection and Modeling for Recognition - Biometrics Research ...

esult in different recognition approaches. Extensive reviews of approaches to face
recognition were published in 1995 [37], 1999 [31], and in 2000 [38].
A workshop
on face processing in 1985 [91] presented studies of face recognition mainly from the
viewpoint of cognitive psychology. Studies of feature-based face recognition, computer
caricatures, and the use of face surfaces in simulation and animation were summarized
in 1992 [49]. In 1997, Uwechue et al. [92] gave details of face recognition based on highorder
neural networks using 2D face patterns. In 1998, lectures on face recognition
using 2D face patterns were presented from theory to applications [36]. In 1999,
Hallinan et al. [93] described face recognition using both the statistical models for
2D face patterns and the 3D face surfaces. In 2000, Gong et al. [94] emphasized
the statistical learning methods in holistic recognition approaches and discussed face
recognition from the viewpoint of dynamic vision.
The above studies show that the face recognition techniques, especially holistic
methods based on the statistical pattern theory, have greatly advanced over the past
ten years. Face recognition systems (e.g., FaceIt [1] and FaceSnap [2]) are being used
in video surveillance and security monitoring applications. However, more reliable
and robust techniques for face recognition as well as detection are required for several
applications. Except for the recognition applications based on static frontal images
that are taken under well-controlled environments (e.g., indexing and searching large
image database of drivers for issuing driving licenses), the main challenge in face
recognition is to be able to deal with the high degree of variability in human face
images. The sources of variations include inter-subject variations (distinctiveness of
individual appearance) and intra-subject variations (in 3D pose, facial expression,
39