【摘要】 串匹配技术是多边形形状识别与分类的一种常用方法,它主要是寻找从一个串到另一个串的最小代价变换,计算变换的相似测度,最终完成目标对象与模型对象的匹配。但是,由于图象分割中存在的分割不一致现象,对属性串匹配的代价因子计算影响很大,实际上也严重限制了串匹配技术的应用。本文研究依据顶点演化规则,首先对目标物体施以两种演化变换——合并与生长,将目标对象递归蜕化成与模型边数相同的近似多边形,再进行等长循环属性串的匹配,由两阶段的变换距离来决定是否匹配。该算法不仅克服了分割不一致现象的影响,而且在计算的复杂度等方面也比常规串匹配算法简单。更多还原

【Abstract】 String-matching is a common approach to recognizing and classifying polygons.It mainly looks for a transformation with minimum cost from one string to another, and computes the dissimilarity measurements between two polygons to totally match each other. Because of the inconsistency in image segmentation, the computation of matching cost for attribute strings is greatly affected, and thus the application of string matching is heavily limited. Based on the rules of vertex evolution, we first impose two evolution transformations of merge and grow on the target objects,turn them into an approximated polygon with the same number of edges as the model,perform an equal loop attribute string matching, and decide the matching by the two-step transformation distances. The algorithm not only overcomes the influence of inconsistency, but also is simpler than conventional string matching algorithms in computation complexity.更多还原