基于最大类间可分离性的轮廓线分割

CONTOUR SEGMENTATION BASED ON MAXIMAL BETWEEN-CLASS SEPARABILITY

对轮廓线上每点的前后两个局部段进行局部直线拟合 ,在此基础上计算这两个局部直线段的类间可分离度 ,证明了它就是轮廓线点是否为一优势点的对数似然比 ,并把轮廓线上的优势点定义为具有局部最大类间可分离度的轮廓点 .基于此概念 ,开发了用直线多边形良好逼近轮廓线的轮廓线分割算法 ,该算法能精确鲁棒地检测轮廓线优势点 .实验结果表明了该技术的性能是令人满意的 .

The between_class separability of the two lines is computed.By local and separate fitting of two straight lines to two point sets before or after the central a point in the contour.The dominant point is defined as the contour point with a local maximal between_class separability,which is proved to be the logarithmic likelhood function of the statistical test about the hypothesis whether the concerned point is a dominant point of the contour.In terms of the maximum between_class separability has been used to develop an algorithm for robust and accurate segmentation of the contour which can be well fitted by a polygon.The experiments have demonstrated that the algorithm can work well with high accuracy and robustness.