Segmentation of faces in video footage using controlled weights on HSV color

Abstract

Accurate and reliable automatic segmentation of faces in video footages is often hard to succeed, leading instead to laborious and tedious interactive manual segmentation. This paper presents a segmentation method that uses a few controlled sets of the weights on HSV components. First, it is shown that HSV has advantages over RGB or YCbCr when segmenting a face in image in such that a binary pattern reflects as many features of the face as possible. Then, a face detection system is constructed, in which each time a significant scene change is detected segmentation is carried out for the beginning frame of a new scene using a few sets of the weights on HSV components, and resulting patterns are correlated with a typical face pattern. Computer experiments show that the successful detection rate is more than 95 out of 100 faces. © Springer-Verlag 2003.