Face representation using hierarchical model

Facial appearance changes in video sequences, because uncontrolled variations of facial appearances due to illumination, pose, expression and low-resolution. This paper addresses this problem by face representation. Hierarchical model is employed to treat facial image reconstruction non-uniformly corresponding to different facial features. Firstly, we remove the illumination by homomorphic filtering. Then, pose and expression are estimated by calculating optical flow, while pose and expression are compensated. Thirdly, face image is segmented in spatial domain and frequency domain respectively, and six layers are obtained and processed by interpolation, features are extracted from each layer. Finally, hierarchical model is constructed as features from several layers are fused and face image is reconstructed. As a result, uncontrolled variations of facial appearances are handled and super-resolution face image is reconstructed to allow for successful recognition.