摘要
在研究局部极大边界判别式嵌入的基础上,提出加权邻域极大边界判别式嵌入算法。该算法是一种基于流形的特征提取算法,在构建目标函数时采用数据的最优重构系数,能够较好地保留数据的邻域几何结构,且不用计算高维矩阵的逆,克服了特征提取中的小样本问题。在2个通用人脸库上的识别实验结果证明,该算法充分利用了每一个流形的判别信息,在缩小同一类别邻域节点距离的同时增加不同类别邻域节点之间的距离,有效区分了不同的类别,能够获得较好的识别结果。
Based on Local Maximal Margin Discriminant Embedding(LMMDE) method,an algorithm named Weighted Neighborhood Maximum Margin Discriminant Embedding(WNMMDE) is proposed.This algorithm is a feature extraction algorithm based on manifold.It preserves the neighborhood geometry structure of the data while constructing objective function by optimal reconstruction coefficient of data.At the same time,the algorithm does not need to compute the inverse of the high dimension matrix,and it can overcome the small sample problem in feature extraction.Recognition experimental results on two general face image database show that the proposed algorithm makes full use of the discriminant information of each manifold,minimizes the distance between the same class of neighboring nodes as far as possible,increases the distance between different classes of neighboring nodes,which effectively distinguishes different categories,and can get better recognition results.
出处
《计算机工程》
CAS
CSCD
北大核心
2016年第6期167-170,共4页
Computer Engineering
基金
国家自然科学青年基金资助项目(61203143)
沪江基金资助项目(C14002)
关键词
人脸识别
特征提取
极大边界
加权邻域
判别式嵌入
face recognition
characteristic extraction
maximum margin
weighted neighborhood
discriminant embedding