Emerging Topics in Computer Vision

5.3. Tensor Voting in 2D

This section introduces the tensor voting framework in 2D. It begins by describing the original second-order representation and voting of [40]. Since the strictly second-order framework cannot handle structure discontinuities, first-order information has been added to the framework. Now tokens are encoded with second-order tensors and polarity vectors. These tokens propagate their information to their neighbors by casting first- and second-order votes. We demonstrate how all types of votes are derived from the same voting function and how voting fields are generated. Votes are accumulated at every token to infer the type of perceptual structure or structures going through it along with their orientation. Voting from token ...