摘要
针对最大间距准则方法在特征提取中没有考虑原始样本的分布而执行硬分类标准的问题,提出了一种基于分类概率保持的最大间距准则人脸识别方法.首先,计算每个样本的分类概率,并且利用分类概率重新定义了样本的类内和类间散度矩阵;然后利用最大间距准则得到最优投影矩阵;最后将原始样本投影到低维特征空间,保持样本分布信息.在ORL、Yale及FERET人脸数据库上的实验表明,该方法在提高人脸识别率上是有效的.
Without considering the distribution of the original samples and performing hard classification problemin feature extraction of the maximum margin criterion method,maximum margin criterion face feature extractionmethod based on classification probability preserving is proposed. Firstly the classification probability of eachsample is calculated,and classification probability is used to redefine the divergence matrixes of the samples in theclasses and between classes;Then the maximum margin criteria are used to get the optimal projection matrix;Finally the original samples are projectied to low dimensional feature space,the sample distribution informationis keeped. On ORL,Yale and FERET face databases,experimental results show that the method is effective inincreasing human face recognition rate.
出处
《河南科学》
2016年第8期1220-1225,共6页
Henan Science
基金
陕西省自然科学基础研究计划项目(2014JM2-6098)
陕西省教育厅自然科学研究计划项目(2013JK0597)
关键词
分类概率保持
最大间距准则
人脸识别
classification probability preserving
maximum margin criterion
face recognition