This paper proposes a robust algorithm of facial expression synthesis.Decompose a train set into person subspace

expression subspace and feature subspace by employing higher-order singular value decomposition(HOSVD)

and then map them to expression subspace

which can be used to synthesize an expressive image of arbitrary strength

given any frontal facial image.By warping source image to generate the target image rather than by linear-combining the images in train set as usual

this approach not only lessens the train set data storage and the computation complexity

but also enables the system to cope with facial images of arbitrary size

background

illumination condition

coloration or pose.With square interpolation

expression facial images of any assigned strength can be obtained.Experiments prove that this algorithm has higher efficiency and can generate very excellent images.