动态规整核支持向量数据描述及在在线签名验证中的应用研究

Support Vector Data Description with Dynamic Time Alignment Kernel for On-line Signature Verification

根据在线签名自动验证的特点和基于支持向量的数据描述方法(SVDD)在小样本一类分类问题上的优越性,提出动态规整核支持向量数据描述(DTAK-SVDD)算法并基于此构建了签名验证系统,对其中的数据压缩方法等实际问题进行了研究.该方案避免了模板的人为选择并可实现判决阈值的自动确定.以签名过程中的力矢量F、力变化率矢量dF和笔尖轨迹矢量S为特征进行了实验验证.结果表明,该方法判别正确率较高,有实际的应用价值.

Presented a new class of support vector data description by incorporating an idea of dynamic time alignment into the kernel function （DTAK-SVDD）. A novel approach for on-line signature verification based on DTAK-SVDD which can carry out automatic choice of matching template and decision threshold is presented and the details of the approach is investigated. The validation is examined with features of pen-force, pen-force differential and pen-position. Experimental results indicate the effectiveness of the proposed solution.