Super-resolution Reconstruction for Facial Images Based on Local Principal Component Analysis

In order to improve the ghosting effect appearing the reconstruction results which are obtained through applying Principle Component Analysis (PCA) on the whole images, a novel algorithm which is reconstructed through applying PCA on local image patches is proposed. The new method firstly proposes the image patch model with overlapping areas. Then the input low-resolution patches are projected on the sample patches through PCA. And the weights can be obtained. Furthermore, the corresponding high-resolution patches are linearly combined through these weights to output the fusion patches. Now the best results are 16x12 reconstructions with the magnification of 8x8. Experimentations show that our method can reconstruct ultra-low resolution faces of 8x6 pixels with the magnification of 16x16, and the similarity with the original high-resolution images is higher.