Adaptive techniques with polynomial models for segmentation, approximation and analysis of faces in video sequences