A Flexible Approach to 3d Reconstruction from Single Images

Traditional image-based 3D reconstruction methods use multiple images to extract 3D geometry. However, it is not always possible to obtain such images, for example when reconstructing destroyed structures using existing photographs or paintings with proper perspective (figure 1), and reconstructing objects without actually visiting the site using images from the web or postcards (figure 2). Even when multiple images are possible, parts of the scene appear in only one image due to occlusions and/or lack of features to match between images. Methods for 3D reconstruction from a single image do exist (e.g. [1] and [2]). We present a new method that is more accurate and more flexible so that it can model a wider variety of sites and structures than existing methods. Using this approach, we reconstructed in 3D many destroyed structures using old photographs and paintings. Sites all over the world have been reconstructed from tourist pictures, web pages, and postcards.