摘要
基于贝叶斯空间的人脸识别算法均假定样本空间满足高斯分布,实际上样本空间很复杂,不一定能满足高斯分布。提出一种新的在贝叶斯空间进行人脸识别的算法,该算法通过设定图像灰度级的阈值,统计其出现频率,计算其类条件概率密度,利用贝叶斯公式求后验概率。该方法克服了传统贝叶斯方法难求类内和类间协方差矩阵的缺点,简单易用。实验结果证明,该方法具有可行性,识别率高于传统的基于代数的人脸识别算法(PCA、LDA和PCA+LDA)。
Traditional face recognition algorithm based on Bayesian space is assumed to meet the Gaussian distribution of the sample space.In fact the sample space is very complicated and does necessarily satisfy the Gaussian distribution.This paper presents a new face recognition algorithm in the Bayesian space.By setting the image gray-level threshold and counting the frequency of the pix and calculate the class conditional probability density,this algorithm gets posterior probability by the Bayesian formula.This method overcomes the shortcomings of traditional Bayesian approach hard to find within-class and between-class covariance matrix,and it is easy to use.Experimental results show that the method is feasible,superior to the traditional algebra-based face recognition algorithms(PCA,LDA and PCA + LDA).
出处
《计算机工程》
CAS
CSCD
北大核心
2011年第5期219-220,223,共3页
Computer Engineering
基金
国家自然科学基金资助项目(60372049)
江西省科技计划青年基金资助项目(GJJ09412)
关键词
贝叶斯子空间
人脸识别
后验概率
Bayesian subspace
face recognition
posterior probability