Super-Resolution of Face Images Based on Adaptive Markov Network

Abstract

Adopting a patch-based Markov network as the fundamental mechanism, we first propose a patch-position constraint operation for searching matched patches in the training dataset to increase the probability value of observation function. For the hidden nodes, based on the first advantage and discovering that horizontal features of the face is more significant than vertical features visually, we create a local compatibility-checking algorithm which uses the most compatible neighboring patches along horizontal dimension of the face to synthesize the super-resolved outcome. Experiments demonstrate the effectiveness of the proposed algorithm.