As said before, the first part of the identified solution reduces the area that is subject
to the later detection, to small regions. The detection area is constrained in a “topdown”
manner to so-called regions of interest (ROI), here, first the face, then eye,mouth and nose. The ”region-detection” makes use of the Haar-Like Feature Extraction with AdaBoost (see e.g. [3]) at the beginning. The Haar-Like Feature Extraction is an object recognition technique which uses weak features to detect domain objects. With the learning algorithm AdaBoost a classifier is constructed for the detector.