An improved morphable model for 3D face synthesis

We present several improvements on the conventional morphable model. Despite having realistic results and modeling automatically in 3D face synthesis, the model construction depends on the unstable optical flow algorithm; the model is not fitting for complex illumination and the model matching procedure is time consuming. To improve the morphable model for 3D face synthesis, several measures are taken in this work. Firstly, the mesh resampling method is used to overcome the key problem of model construction, the pixel-to-pixel alignments of prototypic faces. Secondly, based on the mesh resampling data we construct a multi-resolution model to accelerate the model matching procedure and improve the stability of matching convergence. Thirdly, a multi-lights model is proposed to evaluate or simulate the illumination of the input facial image. The experimental results show that these improvements of the morphable model have good performance in 3D face synthesis.