A Study for Inductive Learning of Structural or Topological Features of Images by Genetic Algorithm

本课题取得了以下成果：(1)设计了一种基于遗传算法的闭合轮廓分割算法，并实现了一个原型系统。(2)对部分字体集进行了实验，取得了成功的结果。(3)根据分类信息，在自动分类实验中取得了成功的结果。(4)设计了每类基元特征的选取算法，并进行了充分的实验。(5)利用上述训练特征集，从根本上考虑了自动生成识别器程序的可能性。(8)考虑了一种基于提取的原始形状特征的形状匹配算法。

The following results have been obtained in the term of the project :(1) We have designed a segmentation algorithm of closed contours based on genetic algorithm, and then implemented a prototype system.(2) We have obtained a successful results with some sets of character fonts.(3) Depending on classification information, we have had successful results in the experimentation of automatic classification.(4) We have designed a selection algorithm of primitive features of each class, and then had sufficient experimental results.(5) Using the training feature set abovementioned, we got sufficient results.(6) We have considered a representation scheme of general primitive shapes.(7) We have considered fundamentally a possibility of automatic generation of recognizer programs.(8) We have considered a shape matching algorithm based on the extracted primitive shape features.