基于改进LPP和ECOC-SVMS的离线签名识别方法

提出一种基于改进LPP和ECOC-SVMS的离线签名识别方法。针对预处理后的签名图像，选择多种有效特征构建高维特征向量，引入一种改进的保局投影方法进行特征提取并同时实现高效降维；签名识别方面，使用基于Hadamard纠错编码方法的ECOC支持向量机多类分类方法，并引入近似概率对ECOC解码进行改进，以提升多类分类器的性能。实验结果表明此方法的可行性和有效性。

Method of Off-line Signature Recognition Based on Improved LPP and ECOC-SVMS

A method of off-line signature recognition based on locality preserving projection(LPP) and Error Correcting Output Code support vector machine(ECOC-SVMS) is proposed. After selecting multiple features from preprocessed signature images, high dimensionality feature vectors are constructed. Then, an improved LPP method is used to extract effect features and reduce dimensionality. A multi-classification classifier based on Hadamard code ECOC-SVMS is used to deal with signature recognition problem. A proximate probability output of SVMS is employed to improve the decoding processing of ECOC framework to enhance the performance of multi-classification. The experiment result shows that the proposed method is feasible and effective.