摘要
针对实际人脸图像含有的噪声模型常常表现出的非高斯特性,该非高斯特性具有较厚重的拖尾现象,提出一种基于多元混合高斯分布的多分类人脸识别方法。该方法将多元混合高斯分布、核函数、概率密度函数估计中的参数估计以及贝叶斯理论结合起来,能对含有重尾噪声的人脸图像有较高的识别率。用ORL标准人脸库进行验证,实验结果表明了可行性。
Noise models often show the non-Gaussian characteristics in the actual face image, this non-Gaussian characteris- tics had a thick tail phenomenon, the paper proposed a kind of face recognition method classification based on muhivariate Gaussian mixture distribution. This method combined multivariate mixed Gaussian distribution and kernel function and parame- ter estimation of probability density function estimation and Bayesian theory, about face image containing heavy tail noise had high recognition rate. Use standard ORL face library verification, the experimental results show its feasibility.
出处
《计算机应用研究》
CSCD
北大核心
2013年第9期2868-2871,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61272210)
江苏省自然科学基金资助项目(BK2012552)
关键词
重尾噪声
多元混合高斯分布
参数估计
核函数
贝叶斯理论
heavy-tailed noise
multivariate mixed Gaussian distribution
parameter estimation
kernel function
Bayesian theory