一种有效的平面曲线关键点检测新方法

平面曲线的关键点检测可用于数字曲线的直线段逼近,因此在很多图像识别、测量及分析的场合均有重要应用.由于一般的数字直线均为若干连续的水平、垂直、±45°夹角直线段合成,称之为四个基本方向,并将每个基本方向的最大直线段长度定义为其方向象素长度(directional pixel length,DPL).提出了四个定理,证明了在相互垂直方向上的DPL不会同时超过1个象素,而45°夹角方向上DPL不会同时超过2个象素.在此基础上,提出并设计了一种全自动关键点检测算法.通过和现有典型算法的实验对比,验证了该方法在减少关键点数和时空开销方面的显著优势.

A New Effective Method on Critical Point Detection of Planar Curves

Critical point detection (CPD) of planar curves can be well applied in approaching of curves using poly-lines,which has very important applications in image recognition,measurement and analysis.As digital lines are commonly composed by several continuous short lines,which are horizontal (k is 0),vertical (k is 1) and ±45°(k is ±1) lines.We call these four lines basic lines and define the maximum length of every basic line as corresponded directional pixel length (DPL),which are further denoted as Lx,Ly,L+ and L-.Furthermore,four theorems on DPL are proposed,based on which,an automatic algorithm for CPD is also presented.The comparisons of algorithms from Ansari-Huang,Zhu-Chirlian and ours was showed.The satisfactory experimental results have demonstrated that our algorithm has apparent advantages in reducing critical point number and spatial-temporal spending.