Subspace Learning Based Facial Feature Analysis And Application

Human face plays an important role in human communication, facial expression shows the complex transmission of human emotions and feelings. In this paper, based on subspace learning and feature analysis, we study the key algorithms of 3D facial expression recognition and facial photo synthesis from photo.This paper proposes facial expression recognition method based on 3D scan data. We combine both the feature selection and feature fusion technique to dig the features which hold the strongest discriminative ability to handle the recognition problem. The synthesized feature fused geometrically localized facial features(GLF) and surface curvature features(SCF) together describe the facial shape more comprehensively. Extensive experiments are carried out on the BU-3DFE database and our results outperform the results presented in the previous work using single features.To synthesis facial photo from sketch, we study the relationship between sketch space and photo space. Before the synthesis, the dictionaries for sketch and photo patches are learned from training dataset. Then we use local coordinate coding to capture the nonlinear relationship between sketch and photo space, at the same time, the simulated annealing algorithm is employed to iteratively update the patches to optimize the nonparametric Markov network by decreasing global energy cost. The convincingly results validate the effectiveness of the proposed method compared to the other existing methods.Through the subspace learning and feature analysis, we can not only compress and filter the redundant and high-dimensional feature space, but also extende facial feature space. The subspace based method can be applied to many research areas of face study to optimize the results of traditional methods.