Abstarct: An accurate reservoir characterization is a crucial task for the development of quantitative geological models and reservoir simulation. In the present thesis, the advantages of thin section images were employed to predict porosity and permeability of carbonate reservoir rocks. Two approaches were used to predict the permeability. First, several static characteristics of thin section images, including distribution of pore spaces, cement percentage, geometrical shape and complexity of pore spaces, were extracted and used as inputs of classification models, including decision trees, discriminate function, support vector machine, K-nearest neighbors and two ensemble algorithms, named boosting and bagging strategies. The models classified the reservoir rock intervals into four groups qualitatively. The obtained results showed that the bagging decision tree model delivering the best performance among the models and improveing the accuracy of simple models up to 7.7% compared with the best single classifier. In the second approach, the thin section images were used as training images of two porous media reconstruction algorithms, named Markov Chain Monte Carlo (MCMC) and Cross Correlation Simulation (CCSIM) algorithm. The output of these models is a 3D digital model, which are suitable to simulate the fluid flow simulation using Lattice Boltzmann Method (LMB). The results introduced the CCSIM as an unreliable algorithm for reconstruction of carbonate media inspit of its acceptable performance in sandstone samples. Lack of connection and relationship of pore spaces in the third dimension are the main drawback of CCSIM algorithm. From the other hand, the MCMC algorithm reconstructed the porous media accurately. The results of fluid flow simulation in the reconstructed models of Kangan and Dalan formations confirm this fact. In addition, the strategy of pore space differentiation and development of sub-models, named micro-model, meso-model and macro model for each thin section facilitated the study of different pore space contribution in the absolute permeability. The porous media of five samples taken from Kangan and Dalan formations were reconstructed using the MCMC algorithm. The fluid flow was simulated in the reconstructed samples and the absolute permeability was estimated. The results show: 1- the macro-porosity has the highest contribution in the absolute permeability; 2- High abundance of micro-porosity does not guarantee the high permeability. The reconstructed sub-models for each sample were integrated through a superposition operator. Fluid flow simulation in the superposition model shows that micro pores improve the throat distribution of macro pores. The mean values of absolute error of horizontal permeability versus the laboratory measurements for the superposition and simple models are 12.77 and 18.89 mD, respectively. In addition, the mean values of absolute error for the vertical permeability in the aforementioned models are 4.69 and 6.92 mD, respectively.