The characteristics of low porosity，low permeability and strong heterogeneity in tight sandstone reservoirs make the pore structure of rocks complicated. Intensive study on pore structure parameters is of great significance to improve oil and gas recovery and reservoir development for the low permeability reservoirs. The rock thin section analysis is the most basic way to analyze the pore structure. The method is a manual approach，which has the shortage of large random error and time-consuming. In order to fully exploit the abundant information of the pore structure in the rock thin section，six samples from Chang9 oil layer in Jiyuan Oilfield were selected to obtain pore structure parameters based on the pore extracted from binarization image with high SNR of pore-skeleton. Through the Bayesian classification method based on the RGB color space model，the parameters such as pore，pore shape factors and porosity were obtained through statistical methods. The calculated porosity by method of image analysis based on Bayesian classification is in linear agreement with the measured porosity and permeability. At the same time，we can concluded that there is a high correlation coefficient（above 0.8）between the pore structure parameters obtained by the method above and the mercury penetration. The calculated results show that this method could obtain more accurate pore structure parameters，which improves the efficiency of rock image analysis. This method provides an effective approach for the characterization of the pore structure in the tight sandstone reservoir.