The heterogeneous pore space of porous media strongly affects the storage and migration of oil and gas in the reservoir. To accurately reproduce complex pore structure validated with experiment data, a hybrid method which combines the CCSIM-TSS method (the cross correlation based simulation-three step sampling method) with the optimization method is proposed to reconstruct stochastically 3D models of the heterogeneous porous media based on 2D images and some constrain conditions. Moreover, the new system in the hybrid method is generated by exchanging the objective points and boundary points of pore and matrix based on two algorithms. Then the hybrid method are tested on the two typical samples (carbonate rock, tight sandstone samples with micro-nanopores). Quantitative comparison is made by computing various statistical and petrophysical properties for the reconstructed models, as well as the original samples. It was found that the hybrid method generates an ensemble of stochastic realization that honor the constrain conditions (permeability data, pore-size distributions and the connectivity). Generating realization of tight sandstones that match the experimental data based on 2D SEM image shows that this method may be then used for more accurate characterization of tight sandstones oil reservoir and analysis of their pore network.