一种基于多项式拟合的人脸识别验证算法分析

Analysis of face recognition verification algorithm based on polynomial fitting

在本文中提出了一种用于人脸识别的验证算法,该算法可以在决策错误的情况下,仍然能识别出两个人脸数据的不匹配对。该系统采用的算法是通过之前未看到的两个面之间的累积绝对差的导数来执行的。与传统的多维距离测量不同,该算法还考虑了累积绝对差值相对于高斯分量的增加趋势。此外,来自脸部特征数据包而得到的人脸识别数据集(LFW,Labeled Face in the Wild)的高斯混合模型,也广泛适用于多种生物外形识别系统。

In this paper,a verification algorithm for face recognition is proposed.The algorithm can still recognize the mismatches of two face data even if the decision is wrong.The algorithm used in the system is implemented by the derivative of the cumulative absolute difference between the two faces that was not seen before.Different from the traditional multi-dimensional distance measurement,the algorithm also considers the increasing trend of cumulative absolute difference relative to Gaussian component.In addition,the Gaussian mixture model of face recognition data set (LFW,labeled face in the wild) derived from facial feature data packet is also widely used in many biometric shape recognition systems.