Facial feature extraction from a video sequence using independent component analysis (ICA)

This paper presents methods for face localization and facial feature extraction from a video image sequence to aid in constructing a 3D face model for animation, a newly added feature in MPEG-4. Face localization is achieved by using human skin color filtering in YCbCr subspace with other relevant facial information. The method of independent component analysis (ICA) is applied to the localized facial images to identify main facial components such as eyes and mouth in the video sequence.