摘要
本文将从多个侧面系统研究独立分类器集成的理论,包括:给出分类器独立性的定义,推导独立分类器的集成方法,从信息论的角度考察这种集成提高识别率的机理,讨论分类器独立性的判别准则。另外,本文将分析概率估计误差对这种集成造成的影响,提出相应的改进算法。本文的理论和方法在脱机手写汉字识别中得到了成功的应用。
In this paper a systematic research is conducted on the combination of multiple independend classifiers. The definition is proposed, the combination method is derived, the effect resulted from estimation error of probabilities is analyzed and a related modification is presented. Practice in off-line handwritten Chinese character recognition proves the effectiveness of the proposed concept and methods.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
1998年第4期403-411,共9页
Pattern Recognition and Artificial Intelligence
关键词
分类器集成
模式识别
字符识别
汉字信息处理
Chinese Character Recogition, Classifier Combination, Classifier Indepenndence, Posteriori Probabilities
