Efficient face hallucination by using shape and texture dependency

In face hallucination problem it is critical to synthesize the correct facial details. General tendency is using textural priors either in parametric or non-parametric fashion. Though a certain amount of details could be obtained, limitations in texture priors force the utilization of shape information in addition to texture. As in Resolution Aware Fitting (RAF) algorithm, better results can be obtained by using shape and texture information together, but independently. However it is known that for constraint domain images shape and texture components have statistical dependency. The utilization of this dependency relation together with component priors would regularize the solution more. In this work a fast synthesis based approach is proposed for the face hallucination problem by incorporating shape information, texture information and the statistical dependency of the image components simultaneously. Experiments show that more accurate reconstructions can be obtained with less computational effort.