基于置信度的手写体数字识别多分类器动态组合

Multiple Classifiers Dynamical Combination of Script Character Recognition Based on Confidence

多分类器组合利用不同分类器、不同特征之间的互补性，提高了组合分类器的识别率。传统的组合方法里，各分类器在组合中所承担的角色是固定的，而实际应用中，对于不同的测试样本，每个分类器识别结果的可信度是不同的。该文根据分类器置信度理论，提出了各类别的置信度。用测试样本自身的置信度信息实现分类器的动态组合，并把这种动态组合方法具体应用到手写体数字的识别。这种方法还可以在不影响已有数据的情况下添加新的分类器进行组合。

Multiple classifiers combination makes use of the complementarities of different classifiers and different characters to improve recognition correctness.In traditional combination methods, each classifier is taken on as a fixed role. But in fact the reliability of very classifier is different for different testing pattern. This paper proposes each class confidence based on classifier confidence theory, realizes a dynamical combination method with each class confidence, and applies this method to script character recognition. New classifiers can be appended by using the method without affecting former training data.