Investigations of Tensor Voting Modeling

Tensor voting (TV) is a method for inferring geometric structures from sparse, irregular and possibly noisy input. It was initially proposed by Guy and Medioni [Guy96] and has been applied to several computer vision applications. TV generates a dense output field in a domain by dispersing information associated with sparse input tokens. In 3-D this implies that a surface can be generated from a set of input data, giving tensor voting a potential application in surface modeling. We study the tensor voting methodology in a modeling context by implementing a simple 3-D modeling tool. The user creates a surface from a set of points and normals. The user may interact with these tokens in order to modify the surface. We describe the results of our investigation.