基于多特征的多分类器融合决策印鉴识别

印鉴真伪鉴别的难点在于由于人工加盖印鉴时,用力的不同,媒介的不同会导致真印鉴的差异与伪造印鉴的差异很难提出一个标准进行自动区别。要求识别系统同时具备同类印鉴的鲁棒性及异类印鉴的敏感性。针对这一难点,该文提出了一种基于多特征的多分类器融合决策算法,根据多类特征以及多种分类器的不同特性,获得对真伪印鉴的鉴别。采用Gabor滤波器获得频率特征,采用差图像获得结构特征,采用原图像和极坐标图像的奇异值获得不变量特征。采用支持向量机(SupportVectorMachine,SVM)和Fisher分类器进行融合决策对印鉴进行真伪鉴别。实验表明,该文方法具有很高的真伪鉴别能力。

Seal Imprint Verification Based on Multi Supplemental Features of Multi-Classifier Fusion Decision

This paper presents a multi-classifier decision fusion algorithm based multiple supplemental features extraction to improve the efficiency of seal imprint verification.Genuine and forgery seal imprints are classified according to effective supplement of multiple features.Structural features are extracted from diff-image.Multiple channels Gabor filters are introduced to extract frequency domain features;SVD is applied to polar coordinates image to obtain invariant algebraic features.Fusion of classifiers like Fisher Discriminant and SVM is adopted to implement final classification decision.Experimental results show that high recognition rate can be achieved.