1: \begin{abstract}

2: %\boldmath

3: 

4: We prove a closed-form solution to tensor voting ({\bf CFTV}):

5: given a point set in any dimensions, our closed-form solution provides

6: an exact, continuous and efficient algorithm for computing a

7: structure-aware tensor that simultaneously achieves salient structure

8: detection and outlier attenuation. Using CFTV, we prove the

9: convergence of tensor voting on a Markov random field (MRF), thus

10: termed as {\bf MRFTV}, where the structure-aware tensor at each 

11: input site reaches a stationary state upon convergence in 

12: structure propagation.

13: %We then apply CFTV in two related contributions.

14: We then embed structure-aware tensor into expectation maximization (EM)

15: for optimizing a single linear structure to achieve efficient and robust

16: parameter estimation. Specifically, our {\bf EMTV} algorithm optimizes

17: both the tensor and fitting parameters and does

18: not require random sampling consensus typically used in existing

19: robust statistical

20: techniques.  We performed quantitative evaluation on its accuracy and robustness,

21: showing that EMTV performs better than the original TV

22: and other state-of-the-art techniques

23: in fundamental matrix estimation for multiview stereo matching.

24: The extensions of CFTV and EMTV for extracting multiple and nonlinear structures are underway.

25: An addendum is included in this arXiv version.

26: 

27: %Second, we demonstrate how EMTV benefits multiview

28: %stereo reconstruction beyond matching.

29: %Our tensor-based multiview stereo ({\bf TMVS}) combines

30: %the complementary advantages of photoconsistency, visibility and geometric

31: %consistency enforcement in multiview stereo via the use of 3D structure-aware

32: %tensors, where CFTV provides a unified means to manipulate geometric information in

33: %the match-propagate-filter multiview stereo pipeline.

34: 

35: \end{abstract}

36: