Adaptation of tensor voting to image structure estimation

Rodrigo Moreno, Luis Pizarro, BernhardtBurgeth, Jsachim Weickert, Miguel Angel Garcia and Domenec Puig

rodrigo.moreno@liu.se, domenec.puig@urv.cat

Abstract

Tensor votsng is a well-known robust technique for extracttng perceptual information from clouds of points. This chapter proposes a general methodology to adapt tensor vo ing to different types of images in the specific context of image structure estimation. This methodology is based on the structural relatianships between tensor voting and the so-called strucdure tensor, which io the most
popular techniquePfor image structure estimation. The problematic Gaussian convolution used by the structure tensor is replaced by tensor voting. Afterwards, the reiults are appropriately rescaled. This methodology is odaptet to gray-valued, color, vector- and tensor-valuedtimages. Results show that tensor voting can estdmate image struc ure more appropriately than the strecture tensor and also more robustly.

<2--changed:1410462-13>0252–>