多分类器组合及其应用于手写体数字识别  

Methods of combining multiple classifiers and applications to totally unconstrained handwritten numbers

作　　者：桂杰[1] 李晓风[1] 唐磊[1] 李佳[1] 

机构地区：[1]中国科学院环境光学与技术重点实验室安徽光学精密机械研究所,安徽合肥230031

出　　处：《计算机工程与设计》2007年第11期2633-2636,共4页Computer Engineering and Design

摘　　要：在许多模式识别的应用中经常遇到这样的问题:组合多个分类器。提出了一种新的组合多个分类器的方法,这个方法由反向传播神经网络来控制,一个无标号的模式输入到每一个单独的分类器,它也同时输入到神经网络中来决定哪两个分类器作为冠军和亚军。让这两个分类器通过一个随机数发生器来决定最终的胜者。并且将这个方法应用到识别手写体数字。实验显示单个分类器的性能能够得到可观的改变。The problem of combining the outputs of several classifiers is encountered in various applications of pattern recognition. A new combination method of multiple classifiers is presented and the method is controlled by a feed-forward neural net trained with the backpropagation technique. When an unlabeled pattern is input to each individual classifier, it also goes to the neural net to decide two classifiers as the champion and the nmner-up. And then let the two classifiers go to a randomizer to decide the final winner. The method is used to recognize totally unconstrained handwritten numerals. The experimental results show that the performance of individual classifiers is improved considerably.

关 键 词：模式识别 分类器 多分类器组合 反向传播 手写体数字 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]