Junction Inference and Classification for Figure Completion using Tensor Voting

We address the issues associated with figure completion, a perceptual grouping task. Endpoints and junctions play a critical role in contour completion by the human visual system and should be an integral part of a computational process that attempts to emulate human perception. A significant body of evidence in the psychology literature points to two types of completion, modal (or orthogonal) and amodal (or parallel). We provide a computational framework which implements both types of completion and integrates a fully automatic decision making mechanism for selecting between them. It proceeds directly from tokens or binary image input, infers descriptions in terms of overlapping layers and labels junctions as T, L and endpoints. It is based on first and second order tensor voting, which facilitate the propagation of local support among tokens. The addition of first order information to the original framework is crucial, since it makes the inference of endpoints and the labeling of junctions possible. We illustrate the approach on several classical inputs, producing interpretations consistent with those of the human visual system.