定向邻域扩展轮廓预估计算法

Contour Estimate Algorithm Based on Fixed Orientation Neighborhood Expansion

为了提高传统Snake面部轮廓检测效率,在Shakable Snake算法的基础上,提出了搜索方向固定的邻域扩展算法.对所搜索到的节点利用动态阀值最小距离聚类法进行分类和修正,使无效节点避开噪声点和噪声边界集中的区域以得到更合理的结果.提出了曲线外推的失效节点修正算法和简化曲率约束准则,以判别修正杂散失效节点.实验得到的结果和感兴趣轮廓重合性在5个像素之内,证明了所提出算法的鲁棒性和有效性,为后续Snake寻找精确边界提供了理想的初始位置.

In order to enhance detection efficiency of traditional Snakes,a fixed orientation neighborhood expansion algorithm was proposed on the basis of Shakable Snake to search possible image contour points.A dynamic threshold minimal distance cluster method was also proposed to judge inappropriate Snake points and revise them to correct location.This method ensured the neighborhood expansion algorithm to find more accurate Snake points.The author also proposed a curve extrapolation algorithm to revise inappropriate Snake points and a new curvature limitation rule to determine invalid points.The examination showed that most of final Snake points were located on the interested contour under different initiate parameters,the estimate error was less than 5 pixels,and the results indicated that the proposed algorithm was of high efficiency.