Chapter 19 - Near real-time robust face and facial-feature detection with information-based maximum discrimination

Abstract

The chapter describes a learning technique to carry out a multiscale search with a face classifier, regarding the face and facial-feature detection algorithm. All face-analysis algorithms start with detecting the presence of faces and accurately tracking their location in complex environments. These algorithms are particularly helpful to interactive systems because they provide the machine with the sense of awareness of the presence of users, thereby triggering alarms, starting applications, initializing systems, etc. This chapter describes a computer-vision and pattern-recognition algorithm for face and facial-feature detection with applications to human-computer intelligent interface. It explains solutions to face and facial-feature detection and discusses issues regarding the implementation and integration of these algorithms into a fully automatic, near real-time face, and facial-feature detection system. It also illustrates important tables and graphs related to the face detection rate. The two image databases referred are the face-recognition technology (FERET) database and Carnegie Mellon University /Massachusetts Institute of Technology (CMU/MIT) Database. Face detection report problems while detecting faces of people wearing very reflective glasses; and ’like other systems it can only handle small degree of in-plane rotation. Also, a relatively large rotation of the user's head degrades the detection results.