Various types of porous media materials inherently contain pore structures of different scales, ranging from nanoscale to millimeter scale. Due to the limitations of the existing imaging technology, it is challenging and intractable for any single method to obtain and characterize the multiscale pore structure features of porous media accurately and comprehensively. To address the issue, according to the inherent logical correspondence and mutual conversion relationship between different‐scale image pixels, we propose an improved universal fusion algorithm for constructing three‐dimensional (3D) multiscale porous media. We successfully applied this algorithm to the construction of multiscale pore structure model of carbonate rock, sandstone and coal, and subsequently made quantitative extractions and characterizations of their pore structure characteristics. In addition, the finite element method (FEM) was used to calculate their absolute permeability. The results show that the improved fusion algorithm can effectively solve the problem of pore bias of the existing algorithm, which reduces the porosity of the multiscale model to a certain extent, while maintaining good pore interconnectivity. Besides, the multiscale model obtained by the improved fusion algorithm has a wider pore size distribution interval than that of the existing algorithm, and the absolute permeability of the former, computed using the FEM, is closer to the laboratory‐measured value than that of the latter.