DISCRIMINATIVE GRAPHICAL MODEL FOR POROUS MEDIA IMAGE SYNTHESIS

I maging synthesis methods open a new door to help scientists for further
study on porous materials. H igh resolution images are required t o analyze the macroscopic propert
ies of porous media. However, a few degenerated high resolution samples are available because of
constraints, and low resolution measurements (such as MRI images) cannot fully describe the
medium. Computer-aided approaches can help the science of porous media by generating many
artificial high resolution samples using the information of available data. In this paper, a novel
discriminative graphical framework is proposed which statistically models the synthesis problem.
The probability distribution of high resolution image of a porous medium given a low resolution
measurement is modeled by conditional random fields (CRF). A Monte Carlo approach is
proposed to sample the constructed model and to generate high resolution samples. Moreover, a
hierarchical CRF is proposed for gradual synthesis of high resolution porous media images. The
success of the models is shown and compared through several experimental results.