Automatic extraction of facial feature points for MPEG4 videophone applications

This paper addresses the problem of automatically extracting a set of characteristic feature points from images containing a front view of a human face. These feature points include edge-points of the basic characteristics of a face and are extremely useful for the adaptation of 3D head models and the extraction of facial animation parameters of the synthetic/natural hybrid coding (SNHC) part of the MPEG4 image coding standard. The proposed method is based on: (a) an initial segmentation of the chrominance image to extract a rough area of the face, (b) post-processing of the extracted area by standard dilation and erosion operators, (c) geometric region splitting within the facial area and (d) detection of the features within their corresponding regions based on their geometrical characteristics.