Inferring 3d Free-Form Shapes From Complex Contour Drawings

Modeling 3D shapes from a single-view user sketch leverages on user’s ability to draw contours of shapes well and allows the user to draw what he is thinking of directly. While the problem of making the inference from an arbitrary complex sketch remains challenging, algorithms have been proposed to solve it for a number of restricted cases. Earlier systems like Igarashi’s Teddy could handle only simple closed strokes. In this chapter we present a method for inferring smooth 3D shapes from more complex contours containing junctions, in particular, T-points and cusps. We propose an interactive, mixed-initiative process, in which the user draws contours, and the computer makes shape inferences based on the user input and also allows the user to either select a different topological interpretation from the list of suggestions or edit the existing intermediate shape representation of hidden contours. The inference process, originally proposed by Williams, involves three basic steps: computing the shape of hidden contours in the user sketch, creating a topological manifold consistent with the full sketch, and smoothly embedding this abstract manifold to create a final plausible 3D shape. We discuss how this framework can be turned into a practical system and propose novel algorithms for completing hidden contours containing cusps in addition to T-junctions, finding a topological embedding of the abstract manifold created by Williams’ method, and creating a fairly smooth solid shape from the topological embedding.