摘要
为提高手写体笔迹识别的识别率和稳定度,将信息融合理论应用到识别算法中,给出了多分类器融合的结构框图;采用将笔迹的特征向量转化为待识样本(检材)的特征向量与样本库中的样本特征向量的距离值作为神经网络的输入值,将多类识别问题转化为判断是否为同一类的问题;提出利用分类器的先验知识,为每个分类器构造一个混淆矩阵,来标识每个分类器的分类能力。
In order to improve the recognition rate and stability, information mixing is applied in recognition algorithm to give multi - classifier mixing＇s structure drawing. Distance value is calculated between samples which are needed to be recognized and those in sample database. Inputs the distance value to the neural network＇s classifier, and the multi - sort recognition is transformed to distinguish whether the two samples are the same sort. Utilizes the prior knowledge of each classifier to estauish an error matrix for each classifier which can measure classifying capability of each classifier.
出处
《长春理工大学学报（自然科学版）》
2005年第4期93-95,共3页
Journal of Changchun University of Science and Technology(Natural Science Edition)
基金
长春理工大学科研发展基金(YJJ2003-01)
关键词
多分类器融合
距离向量
混淆矩阵
multi-classifier mixing
distance vector
error matrix
