A novel separating strategy for face hallucination

A novel separating strategy is proposed for resolving single face hallucination problem when given an input low resolution face image. First, a local patch-based eigentransformation method that can capture the facial prior is introduced for restoration of the facial structure and zoom the input face image to a medium resolution by using the pairwise patch sets of high resolution and low resolution. Secondly, the fine details of face image generated by the first step is further improved by applying patch-based sparse representation and learning the coupled over-complete patch dictionaries preliminarily. Lastly, the superior of the framework is demonstrated by the high quality of the results of several experiments.