Computationally efficient algorithm for face super-resolution using (2D)2-PCA based prior

Abstract

Super-resolution algorithms typically transform images into 1D vectors and operate on these vectors to obtain a high-resolution image. In this study, the authors first propose a 2D method for super-resolution using a 2D model that treats images as matrices. We then apply this 2D model to the super-resolution of face images. Two-directional two-dimensional principal component analysis (PCA) [(2D) 2 -PCA] is an efficient face representation technique where the images are treated as matrices instead of vectors. We use (2D) 2 -PCA to learn the face subspace and use it as a prior to super-resolve face images. Experimental results show that our approach can reconstruct high quality face images with low computational cost.