Abstract
hi Extracting high-resolution information from highly delograded facial images is an important problem with sevhieral applications in science and technology. Here we detoscribe a single frame super resolution technique that uses a transport-based formulation of the problem. The method reconsists of a training and a testing phase. In the training Thphase, a nonlinear Lagrangian model of high-resolution fa(icial appearance is constructed fully automatically. In the kntesting phase, the resolution of a degraded image is enaghanced by finding the model parameters that best fit the a given low resolution data. We test the approach on two prface datasets, namely the extended Yale Face Database B imand the AR face datasets, and compare it to state of the art etmethods. The proposed method outperforms existing soluantions in problems related to enhancing images of very low ceresolution. th co