基于曲率的尺度空间中的角点增强

角点检测在模式识别、计算机视觉和其它相关领域中占有重要地位．由于无法事先确定应在什么尺度下进行检测，我们很难从平面曲线中准确地提取出角点．本文阐述了通过曲线尺度空间（Curvescalespace）来增强角点的方法．该方法的主要思想是通过曲线的形变（或构造曲线尺度空间）使曲线上的角点随着尺度的增加变得更加突出，以便准确地提取出角点．从广义的几何热扩散方程（GGHF）出发，我们得到了该方法中曲线曲率的演化形式，它揭示了构造基于曲率的尺度空间与形成尺度空间所需符合的准则二者间的本质联系．本文研究GGHF符合尺度空间准则的条件并给出了寻找不同的符合尺度空间准则的曲线演化方法，与此同时，我们给出了角点可被增强的条件．在符合上述条件的前提下，本文给出了一种新的构造尺度空间的方法，该方法符合尺度空间的准则，并能够在去除噪声和不重要的小结构的同时，有效地增强曲线上的重要角点．

CORNER ENHANCEMENT IN CURVATURE SCALE SPACE

In this paper, the problem of corner enhancement via curve scale spaceis addressed. The main idea of the approach is to deform the curve(or construct thecurve scale space) so that the corners are easy distinguished from other structures.Started from the General Geometric Heat Flow equation (GGHF), the evolution ofcurvature in the approach, which is an essential relationship between the construction of curvature scale space and the formularization of the scale space criteria, isderived. Then the conditions under which the GGHF satisfies the proposed criteriaare studied so that many different evolution approaches which have regular scalespace properties can be safely found. The conditions under which the corner can beconsidered as "enhanced" are also proposed. And having all these constraints, anew deformation scheme, which can enhance strong corner, suppress noise andsmall structure while satisfies the scale space criteria, is presented.