应用云模型的字符不确定性定性定量双向认知

Qualitative and Quantitative Cognition of Characters Uncertainties Based on Application Cloud Model

由于自身结构、光照强度和拍摄角度等因素的影响,字符表现出显著的不确定性.对于一个字符,人类可以在大量样本的基础上形成定性认知,并且抵抗干扰,正确识别字符.同时,根据定性认知可以生成一个新的字符样本.虽然以深度学习为代表的机器学习模型已经在字符识别上取得了较高的准确率,然而在字符不确定性认知的内在机理分析上还有待提升.使用云模型理论研究常见字符的不确定性.并且利用自编码器对字符特征的提取能力,研究了字符的定性和定量的双向认知方法.最后,基于云模型中的确定度,提出了两种字符分类算法.实验结果表明,云模型融合自编码器的方法可以更清晰地表示对字符认知的过程.在识别准确率上,深度学习方法高于基于云模型的分类方法,云模型融合自编码器的分类方法要优于云模型分类器方法.

There are plenty of uncertainties in characters because of the influence of structure,light intensity and camera angle.As for a character,the human can form a qualitative cognition on the basis of large amount of samples and they can resist interference and can recognize the characters correctly.Moreover,they could generate a new character specimen according to their qualitative cognition.Although machine learning models represented by deep learning have achieved high accuracy in character recognition tasks,the analysis on intrinsic mechanism of cognition of character uncertainties is still needed to be improved.The cloud model theory is utilized to study the uncertainties of frequently used characters.The extraction capability of the characteristics of the characters by auto-encoder is used to study the qualitative and quantitative cognition method.Finally,two kinds of character classification methods based on degree of certainty in cloud model are proposed.The experimental results show that the cloud model incorporating auto-encoder method could express the cognition process of characters more clearly.In character recognition accuracy,the deep learning based method outperforms the cloud model based classification method,and the classification method of cloud model incorporating auto-encoder method outperforms the single cloud model classifier method.