基于MALS双线性分解的表情合成

Study on the facial expression synthesis is a constructive and creative subject in the field of virtual reality technology. The bilinear model is used for facial expression synthesis by separating the identity factor and the expression factor. The translation procedure in the bilinear model always requires a repetitive computation of matrix inverse operations to reach the identity factor and the expression factor. This computation may be instable when the observation data has the correlation or noisy information. In order to increase the stability of the bilinear model for expression synthesis, the modified iterative least square (MALS) regression is introduced into the translation procedure to stabilize the computation. Experimental results demonstrate the effectiveness of the method.