一种多特征组合与多神经网络分类器集成的手写数字识别新方法

提出了一种多特征组合与多神经网络分类器集成的手写体数字字符识别方法。新方法对结构简单、书写规范的数字字符使用高效的结构分类器实现快速分类;对结构分类器不能准确分类的字符,首先通过提取多个特征,构造不同结构和类型的多个神经网络分类器,然后对多个分类器的识别结果,通过多专家集成方案对手写数字进行最终判决。实验结果表明,提出的方法能有效降低识别错误率,提高识别系统的可靠性。

A New Method for Handwritten Numeral Recognition with Combination of Multi-features and Integration of Multi-neural Network Classifiers

A method for the recognition of the handwritten numerals by combining multi-features and integrating multi-neural network classifiers is presented.For the numerals with simple structures and regular writing,efficient structural classifiers are applied for rapid recognition.For those numerals which cannot be classified accurately with the structural classifiers,multiple statistical features are extracted and combined into high dimensionality features.Then,after dimensionality reduction with principle compone...