Face SuperResolution Method Based on Key Points Layer by Layer

This paper proposed face superresolution (SR) method based on key points layer by layer in order to learn a mapping from low resolution (LR) between high resolution (HR). Considering the importance of rebuilding the details of facial features. This paper trained every mapping function of SR separately based on the key points of face image in processing of this method. Each SR mapping function was iterated to train in order to reduce the difficulty of rebuilding. The paper has proposed the linear and nonlinear mapping function in this method. Linear mapping function was trained by Principal Component Analysis (PCA) and Nonlinear mapping function was trained by AutoEncoder (AE). Using the input data initialized by the method of Bilinear interpolation, the SR method trained each SR mapping function of local patches and superimposed the rebuilt patches on the face image so as to get the final result. This SR method shows efficient in the experiment data set. It has proved to be efficient in processing of facerecognition on ID card.