This paper presents effective algorithms for reconstructing three-dimensional models of specific curve structures from single perspective view images. The proposed method begins with edge detection and filtering from photographs or paintings with fine perspective geometry. Feature lines and corner points are extracted by transferring detected edges to normal distance and normal angle space. Curved segments are also extracted separately. Three mutually orthogonal vanishing points are then calculated automatically using k-means cluster and standard deviation filter on the intersections of extended straight feature lines. Feature points are extracted using a corner point filter with geometry constrains, and their corresponding base points also can be predicted with vanishing points. Three-dimensional coordinates of feature points are computed based on vanishing points using single-view metrology. In addition to straight lines and planar features, two types of curved segments are processed and reconstructed, including cylinder shape and figure-of-rotation structures. Finally, regular and curved models are reconstructed from their shared feature points. Experimental results from computer-simulated images and close-range photographs of buildings indicate that the developed algorithms can successfully extract 3D information and reconstruct 3D building models. The developed algorithms are highly automated, providing an effective and efficient method for reconstructing building models with specific curved structures from single-view images.