Facial Features Segmentation by Model

Deformable models have recently been accepted as a standard technique to segment diierent features in facial images. Despite they give a good approximation of the salient features in a facial image, the resulting shapes of the segmentation process seem somewhat artiicial with respect to the natural feature shapes. In this paper we show that active contour models (in particular, rubber snakes) give more close and natural representation of the detected feature shape. Besides, using snakes for facial segmenta-tion frees us from the problem of determination of the numerous weigths of deformable models. Another advantage of rubber snakes is their reduced computational cost. Our experiments using rubber snakes for segmen-tation of facial snapshots have shown a signiicant improvement compared to deformable models.