||Image-based modeling is the process of converting 2D images of the real world into digital 3D models in computer. Among myriad kinds of objects in the world, man-made buildings are of special importance, since there are a large number of potential applications. This challenging problem has been studied by both academic research and commercial industrial communities. However, all the existing approaches still need plenty of human efforts in order to produce satisfactory results. This thesis presents a state-of-the-art automatic approach that only requires minimal human efforts for image-based building modeling to achieve visual pleasing results. This image-based approach has three steps: reconstruction, segmentation and modeling. The first step is to reconstruct a 3D point cloud from the 2D images. The second step is to segment the 3D point cloud and the 2D images, where each group represents an object or a kind of objects for modeling. The final step is to model each object by structure analysis and regularization. This three-step approach is remarkably robust, because it clearly divides the work into subproblems properly, and conquers each subproblem with strategies according to different objectives to be achieved in each stage. While this approach is also suitable for general image-based modeling of any object, specifical focus is on man-made buildings, where Manhattan-world property presents frequently. This thesis concludes with demonstration of the proposed approach for building modeling in several cities, including Pittsburgh, Minneapolis, Chapel Hill in the United States, and Guangzhou in China.