论文部分内容阅读
针对大类别集分类问题提出了一种新的快速分类方法.引入了基于分组的候选规则,通过冗余分组,将大类别集分成若干独立的子集.组的数量和类别数都是有限的,因此可以充分利用各种信息,单独为每个组设计优化的分类器.以手写汉字识别为例,利用多级学习矢量量化来分别训练全局分类器、组中心以及每个组的细分类器.提供了危险区域的判据,并且结合其他的候选规则来提高边缘样本的识别率.
In this paper, a new fast classification method is proposed to solve the problem of large category set classification, and a group-based candidate rule is introduced to separate the large category sets into several independent subsets by the redundant grouping. The number of groups and the number of categories are limited , So we can make full use of various information to design an optimized classifier for each group.With handwritten Chinese character recognition as an example, we use multilevel learning vector quantization to train the global classifier, group center and the classifier of each group The criteria for hazardous areas are provided, and the recognition rate of edge samples is improved in combination with other candidate rules.