Human Face Super-Resolution Using Principal Component Analysis and Residue Compensation

Abstract

This paper proposes a learning based face super-resolution algorithm. The coefficients of reconstruction are obtained based on principle component analysis (PCA). The preliminary face is generated through these coefficients. Then, all residual patches are respectively obtained using the corresponding patches of the same face regions in the training set. Finally, these residual patches are combined to a global face according to their original positions and this global face is used to compensate the generated image of the first step. Experimental results illustrate that the proposed method can produce higher-quality images than some recent methods.