摘要
为强化鉴别信息的完整性,提升解决小样本问题(SSSP)的能力,该文构造了一种求解具有几何对称性的样本鉴别信息的特征抽取算法。从线性子空间的角度出发,利用人脸的几何对称性,依据奇偶分解原理,在原特征空间生成一组镜像对称基函数。构造一种矩阵变换,求出两个对称基之间的过渡矩阵,并在过渡矩阵空间上求取最优鉴别矢量集。该方法强化了鉴别信息的完整性,对解决SSSP是有效的。在ORL和FERET人脸数据库上的实验结果验证了算法的有效性。
To enhance the integrity of discrimination information and improve the ability of solving small sample size problems（SSSP） ,a feature extraction approach is constructed to solve the sample identification information with geometric symmetry. From the view of linear subspace, a set of mirror symmetrical basis functions are constructed in original space according to the geometric symmetry of face images and the odd-even decomposition theorem. A matrix transform is presented to obtain the transition matrix between the two odd-even basis functions. The optimal discrimination vectors are obtained in the transition matrix space. This method enhances the integrity of discrimination information and solves the SSSP effectively. Experimental results from the ORL and FERET face image databases demonstrate the effectiveness of the proposed method.
出处
《南京理工大学学报》
EI
CAS
CSCD
北大核心
2012年第6期915-918,共4页
Journal of Nanjing University of Science and Technology
基金
中国博士后科学基金(2011M500926)
江苏省自然科学基金(BK2012700
BK2011371)
江苏省博士后科学基金(1102063C)
人工智能四川省重点实验室开放基金重点项目(2012RZY02)
关键词
镜像基函数
过渡矩阵
人脸识别
小样本问题
basis function of mirror symmetry
transition matrix
face recognition
small sample size problem
