Inferring the 3D shape of an object from a single image is an important and difficult problem in computer vision. Since an image is a 2D projection, the process is not invertible without making some assumptions. A number of approaches for inferring 3D shape have been suggested, such as shape from shading, texture, and contour. This paper deals with the shape-from-contour problem for two classes of objects: straight homogeneous generalized cylinders (SHGCs), and planar, right constant cross-section generalized cylinders (PRCGCs). SHGCs are generalized cylinders with a straight axis and cross sections of a fixed shape but varying size; PRCGCs have a planar, but not necessarily straight, axis and a cross-section that is fixed in shape and size.
The approach is based on an analysis of the properties of SHGC and PRCGC objects to derive the types of symmetries that the limb boundaries and cross-sections of these objects produce on the image plane. The method does not require a priori knowledge that the objects in a scene belong to these classes. Constraints on the 3D shape of the objects are formulated based on the symmetries and from the geometry of the projection models, and are used for quantitative shape recovery of surfaces. Results for both classes of objects are given. As indicated in the paper, evaluation of shape from contour results is difficult because there is no real “ground truth”; the only good measure of algorithm performance is a comparison with human performance.
The promising results are based only on synthetic examples. It will be interesting to see whether these techniques are effective on boundaries derived from complex real images. The authors believe that missing contours that arise using real images can be filled in by taking advantage of the properties developed by their methodology.