a) Knowledge-based Top-down Methods (Cont.4)
- Level 1: input image is scanned globally to find, according to a set of rules, all possible candidates of human face with all possible sizes and locations.
Level 2: a second set of face detection rules is established on the basis of a window of 8 x 8 cells over each candidate face. The rules are established on the basis of mosaic images (quartet n = 4 and octet n = 8).