Adaptive Markov random fields for example-based super-resolution of faces

Abstract

Image enhancement of low-resolution images can be done throughmethods such as interpolation, super-resolution using multiplevideo frames, and example-based super-resolution. Example-basedsuper-resolution, in particular, is suited to images that have astrong prior (for those frameworks that work on only a singleimage, it is more like image restoration than traditional,multiframe super-resolution). For example, hallucination andMarkov random field (MRF) methods use examples drawn from the samedomain as the image being enhanced to determine what the missing high-frequency information is likely to be. We proposeto use even stronger prior information by extending MRF-basedsuper- resolution to use adaptive observation and transitionfunctions, that is, to make these functions region-dependent. Weshow with face images how we can adapt the modeling for each imagepatch so as to improve the resolution.