最小风险贝叶斯决策的二值化人脸识别算法  

Face recognition algorithm of binary image by smallest risk Bayesian method

作　　者：曾岳[1,2] 冯大政[1] 付达杰[2] 

机构地区：[1]西安电子科技大学雷达信号国家重点研究所,陕西西安710071 [2]江西财经职业学院信息工程系,江西九江332000

出　　处：《计算机工程与设计》2011年第10期3511-3513,共3页Computer Engineering and Design

基　　金：国家自然科学基金项目(60372049);江西省科技计划青年基金项目(GJJ09412)

摘　　要：提出了一种最小风险贝叶斯决策的二值化人脸识别算法,该算法通过设定图像灰度级的阈值对图像进行二值化,统计其出现的频率,计算其类条件概率密度,根据图像的相似性估计其损失函数,利用贝叶斯公式求最小风险,最后根据最小风险判断其所属类别。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。实验结果表明,该方法具有可行性,比传统的基于代数的人脸识别算法(PCA、LDA和PCA+LDA)的识别率高,并能有效减少相似类的重叠。A face recognition algorithm of binary image by the smallest risk Bayesian method is presented. Firstly by setting the imagegray-level threshold the binary image is gotten, then counting its frequency of the pix and calculating the class conditional probability density, this algorithm gets the smallest risk probability to classify it by the Bayesian formula according to it＇ s the loss function gotten by the Image similarity. This method overcomes the shortcomings of traditional Bayesian approach hard to find within-class and betweenclass covariance matrix. A large number of experiments show that the method is feasible, prior to the traditional algebra-based face recognition algorithms （PCA, LDA and PCA ＋ LDA）, and can reduce effectively the similar classification overlapping.

关 键 词：贝叶斯 人脸识别 后验概率 二值化 最小风险 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]