中文摘要：

序列相似性描述是序列分类的关键,根据序列产生的背景和机理,融合利用具有不同物理意义的特征子模式集合进行序列相似性描述可以改进序列分类的效果.对于在多个特征子模式集合的核变换空间上进行的相似性描述,可利用半定规划方法,在使得分类边界距离最大的意义下对核矩阵相似性描述结果进行优化,从而建立起一种能够融合利用多种意义特征子模式集合的序列分类方法.该方法用于手写签名序列的识别实验,在基准签名数据集上取得了较好的实验结果.

英文摘要：

The measure of similarity between sequences is of great important to sequence classification. The classifying results could be improved if the similarity is measured through different set of feature subsequences that are determined according to the background and mechanism of sequences. Kernel matrix is defined to measure sequence＇s similarity on the space spanned by each set of feature subsequences. The combinational optimization of kernels is solved through semidefinite program, which simultaneously makes the classifying distance maximize, therefore a new method for sequence classification is obtained. The new method is used to verify hand written signatures. Experiments on benchmark database show the signature verification accuracy has been enhanced.