An advanced Bayesian image classification algorithm based on Markov Random Fields will be introduced. The algorithm can be directly applied to segment reconstructed multiphase data of porous materials. This eliminates the need for dual-energy or wet/dry scans and associated image alignment and subtraction analysis that are commonly applied in synchrotron Micro-CT. To speed up segmentation of large 3-D datasets (i.e. 4000 x 4000 x 1000 voxels) the code is implemented in CUDA to utilize GPU parallel processing power. Applications to glass bead and soil data are shown and discussed.