基于矩特征和证据理论的离线签名识别  

Off-line Signature Recognition Based on Moments and Evidence Theory

作　　者：陈刚[1] 李弼程[1] 庄军[1] 刘安斐[1] 

机构地区：[1]信息工程大学信息科学系,郑州450002

出　　处：《计算机工程与应用》2005年第18期57-60,共4页Computer Engineering and Applications

摘　　要：论文首先对几种矩特征在离线签名识别中的性能进行了比较,在此基础上选取矩和一种基于中心矩的形状描述子(SDBCM)作为签名图象的形状特征,据此构造了两个距离权重k-NN分类器对签名图象进行初步识别。然后将两个k-NN分类器的度量层输出作为证据,用一种改进的证据理论合成公式对其进行融合得到最终识别结果。实验结果表明,新的识别方法是有效的。After comparing the performance of several moments in off-line signature recognition,the paper selects the Zernike moments and a shape descriptor based on central moments as the shape features,which feed in two classifiers to implement the elementary recognition.Results of the classifiers in the measurement level are then fused,using an improved combined method based on D-S theory which achieves the final recognition results.Experimental results show that the new recognition method is effective.

关 键 词：离线签名识别 矩特征 k—NN分类器 证据理论 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]