基于人脸和语音的混合型身份认证系统

A multi-channel identity verification system based on face and voice

随着当前电子商务和金融交易领域的发展,用户身份认证的应用变得越来越广泛.结合实际,本文构建了一种基于人脸和语音的混合型身份认证系统,分别提出了基于重建误差分类器的特征脸确认算法和基于高斯混合(Gaussian mixture models,GMM)说话人确认算法;最后在分数层进行融合,提出了基于正交多项式核函数的支持向量机.实验结果表明,该方法在分类、泛化能力和减少支持向量数目方面均取得了良好的效果,最终获得了较小的等误差率.

With the development of E-commerce and financial trade, identity verification has been widely used. A verification system based on face and voice was given. Two algorithms of a reconstruction error classifier based on eigen-faces and speaker verification based on the Gaussian mixture model were realized. Finally, an alternative to the construction of the support vector machine （SVM） kernels from the orthogonal polynomials was presented. Experimental results show that the SVMs with orthogonal polynomial kernels outperform those with traditional kernels in terms of generalization power and less support vectors.