Recently two novel approaches to face detection have been devised
which produce results comparable to well established methods, but
allow near real-time image processing. The first method is due to
Viola and Jones (Viola 2001), the second is due to Froba and
Kullbeck (Froba 2000). This paper applies both methods to the
FGNET video conference data set. The aim is to extract all the
human faces from the video sequence, i.e. approximately 10,000
images, with a minimum number of false positives. The merits of
both algorithms are discussed and the results compared.