Study of a Robust Feature: The Pointwise Lipschitz Regularity

摘要

The aim of this paper is to highlight the relevance in computer vision of the pointwise Lipschitz regularity α∈ℝ. The regularity α gives a measure of the local regularity of the intensity function associated to an image. Known wavelet methods provide an efficient computation of α at contour points of the image. From a theoretical point of view, we study the effect of geometric deformations and other specific transformations applied to the image, showing invariance properties. From a practical point of view, we assess the robustness of the regularity α when the image undergoes various transformations. The results we obtain show the Lipschitz regularity α is a suitable feature for applications in computer vision.