FACIAL MOVEMENT INFORMATION EXTRACTING METHOD BASED ON TENDENCY CONSISTENT-GAUSSIAN PROCESSING LATENT VARIABLE MODEL

摘要

A facial movement information extracting method based on a tendency consistent-Gaussian processing latent variable model. The tendency consistent-Gaussian processing latent variable model is described as follows: (1) forming a Gaussian processing latent variable model objective function based on Markov assumptions and for solving the low-dimensional hidden variable sequence; and (2) adding a tendency-consistent limiting condition to form a tendency consistent-Gaussian processing latent variable mode objective function. The facial movement information extracting method based on a tendency consistent-Gaussian processing latent variable model is described as follows: (1) obtaining, by using a principal component analysis (PCA), a facial sequence hidden variable space initial value for the tendency consistent-Gaussian processing latent variable mode objective function; and (2) solving the hidden variable by using the scale conjugate gradient method, to obtain a low-dimensional hidden variable sequence corresponding to the facial movement sequence. Through the method, while the movement information independent of the identity information is extracted, the difference in the range of hidden space sequence variation caused by different-range facial movement is reserved.