基于特征分布的图象信息抽取

A Hierarchical Fuzzy Recognition Algorithm Based on feature Distribution Analysis

为抽取实时图象中的有效信息，有别于传统的将整个特征域看做一个信源的特征筛选方法，将具有统计意义的特征区间看作一个信息源，计算其熵值，取熵值较小、且类间离散度较大、类内离散度较小的特征区间为有效特征域，每个模式都拥有自己的一组贡献值不等的有效特征域构成其专用特征空间。此算法的有效性在工业流水线上的工件识别系统中得到了较满意的验证。

In this paper,a new feature selection method is presented based on analysis of the feature distribution of the images using information theory.In order to extract more useful information of the feature distributions,we calculate the entropy of the feature intervals instead of the whole fearue space,then Special used Multi dimension Feature Space(SMFS) is constructed, which was comprise of effective feature intervals with small entropy, small inter pattern dispersion and large intra pattern dispersion.Different pattern has its distinctive SMFS.The experiment of the component recognition on the assembly lines using this method shows quite satisfing results.