A robust edge indicator employing nonlocal structure tensor

Abstract

In this paper, we propose a robust edge indicator employing two eigenvalues of nonlocal structure tensor matrix. In our method, a new nonlocal structure tensor is first constructed. This structure tensor is robust to noise, which inherits from nonlocal means algorithm. Furthermore, based on the constructed nonlocal structure tensor, a new and edge indicator is built, which can effectively differentiate a pixel at edge from a pixel in flat region with noise. Moreover, the proposed edge indicator can characteristic the strength of the edges. By experiments, the results of our method is superior to the gradient's results.