摘要
为了提高人脸识别的精度,将随机图理论应用到人脸识别中,提出一种融合Gabor特征的基于随机点积图的算法.首先用Gabor滤波算子提取训练集图像的特征数据并进行预处理,减少特征数据内的冗余性.随机图对人脸训练集图像进行重构,获得节点赋值.之后根据赋值向量夹角余弦值以及节点的类别信息,计算每个节点的惩罚权值.将待识别图像投影到训练集图像随机点积重构得到的子空间中,依据最近邻节点惩罚值进行识别,判断图像所属类别.在ORL人脸库上的实验结果表明,算法在对归一化人脸图像特征数据降维上要优于PCA方法.
In order to improve face recognition accuracy by using random graph theory, this paper presents a algorithm fusing Gabor feature based on random dot product graph. First use Gabor filter to extract Gabor feature from the training set. In order to reduce data redundancy, these features require some pre-treatment. Then access node assignment by reconstructing the training set of face images with random graph. According to the cosine of the assignment vector angle and their category information, calculated penalty value for each node. The identification image is projected to the subspaces which is reconstructed by the random dot product graph, classified based on the nearest neighbor node' s penalty value. Experimentation on the ORL face image database indicate the effectiveness of the algorithm.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第6期1306-1309,共4页
Journal of Chinese Computer Systems
基金
高等学校学科创新引智计划项目(B07033)资助
"核高基"重大项目(2009ZX01028-002-003-005)资助
关键词
人脸识别
流行学习
随机点积图
图嵌入
惩罚值矩阵
face recognition
manifold learning
random dot product graph
graph embedding
penalized matrix