Face hallucination based on pyramid regression strategy

In this paper, we propose a novel face hallucination approach based on the pyramid regression strategy´. A pyramid framework is built which divides the whole face hallucination process into several levels. The output of last level is the input of next one. For the face hallucination at each level, the searching strategy is mixed: globally best-matching and locally best-matching. The former is got from the feature pyramids, and the latter the neighboring patches. Experiments demonstrate that our approach can get better results.