Multiple-view face hallucination by a novel regression analysis in tensor space

In this paper, the novel multiple-view face hallucination method was proposed. This method is reconstructed the high-resolution face images in various poses (normal, up, down, left, and right) from a single low-resolution face image within these poses. There are two steps in our proposed method. In the first step, a high-resolution face image in the same view of the observation is reconstructed by the position-patch face hallucination framework with the improved Locally Linear Embedding (LLE), which the number of neighbours is adaptive. In the second step, the reconstructed image is used to generate the highresolution of the other views by the novel tensor regression technique. The experimental results on the well-known dataset show that the proposed method can achieve the better quality image than the baseline methods.