基于决策树学习的商标库建立与检索

Establishment and Retrieval of Trademark Image Database Based on Decision Tree

本文采用决策树学习的方法解决商标库的建立与检索问题。在系统建立过程中,选择商标的骨架图像的相对矩[2]特征作为原始特征空间,力求获得完整的结构和统计特性;提出了距离差函数(DDF)来为决策树的非终结点选择合适的特征,巧妙地解决了特征空间的选择和降维问题;同时,选择决策树的结构作为商标库的结构,利用ISODATA算法的自动学习识别结果作为启发信息,利用深度优先策略对决策树进行遍历,从而将商标的识别、检索与库的建立统一起来。实验结果证明,此种方法是可行、高效的。

In this paper, decision tree learning is used to establish and retriev e tra demark image database. In order to gain full structural and statistical id entities, relative moments of the trademark skeleton image are us ed to form t he original feature space during system establishing. Distance Difference Functi on is created to select good feature for each non-terminal node. Meanwhile, deci sion tree structure is appointed as the structure of trademark database, which h elps unifying the course of trademark recognition, trademark retrieval and trade mark database establishment. The unifying also dues to the selection of classifi cation result of ISODATA as heuristic information and the depth-first strategy, which reduces the searching cost. Finally, the experiment outcome confirms this way of feasible and highly efficient.