摘要
在模式识别统一信息熵理论的基础上,提出基于信息熵的模式特征选择方法。给出在不同情况下,特征选择条件的具体数学表达式,定义特征性能因子来体现某一个特征对模式类别的表征能力和对不同模式类别的区分能力。将提出的特征选择方法应用于静态手写汉字签名特征描述中,利用描述一个签名时不同特征所表现出来的权重值构成该签名的特征掩膜矩阵。对矩阵中的元素进行处理,获得不同书写者签名之间的整体差异度以及在某一具体特征上所表现出来的差异,从而确立一种静态手写汉字签名特征描述及相似度判定的方法,并通过仿真验证了算法的可行性和有效性。
We present a new method for pattern feature selection based on uniformed information entropy theory in pattern recognition. Specific mathematical expression of feature selection is proposed for different conditions, feature performance factor is also defined to represent the eharacterisation capability and the differentiation ability of a feature on different pattern classes. We apply the proposed feature selection scheme to the features description of an off-line handwritten Chinese signature, and use the weight values manifested in different features when describing a signature to from the feature mask matrix of the signature. By dealing with the matrix elements, the difference degrees of signatures in whole or the difference represented in a specific feature both among different writers can be calculated. Therefore, a method of feature description and similarity determination for off-line handwritten Chinese signatures is established, and the algorithm＇ s feasibility and validity are testified by computer simulation.
出处
《计算机应用与软件》
CSCD
北大核心
2013年第1期99-102,225,共5页
Computer Applications and Software
基金
公安部应用创新计划项目(2009YYCXJLST024)
关键词
信息熵
特征选择
特征性能因子
静态手写汉字签名
Information entropy Feature selection Feature performance factor Off-line handwritten Chinese signatures
