Example-Based Regularization Deployed to Face Hallucination

Regularization plays a vital role in ill-posed problems. A properly chosen regularization can direct the solution toward a better quality outcome. An emerging powerful regularization is one that leans on image examples. In this paper, we propose a novel scheme for face hallucination. We target specially the quality of highly zoomed outputs. Our work bases on the pyramid framework and assigns several high-quality candidate patches for each location in the degraded image. We definite a global MAP penalty function to reject all the problematic examples, and then reconstruct the desired image using the patch which is left after pruning. Experimental results demonstrate that our approach can get better resolution.