摘要
针对现有的局部正切空间算法中存在的问题,文中提出一种基于核变换的特征提取方法——核正交判别局部正切空间对齐算法(KOTSDA).该算法首先利用核方法将人脸图像投影到一个高维非线性空间,提取其非线性信息;然后在目标函数中利用正切空间判别分析算法在保持样本的类内局部几何结构的同时最大化类间差异;最后添加正交约束,得到核正交判别局部正切空间对齐算法.该算法不需要经过PCA降维,有效避免判别信息的丢失,在ORL和Yale人脸库上的实验验证算法有效性.
To address the drawbacks of the local tangent space alignment algorithm, a feature extraction method based on kernel transformation, kernel orthogonal discriminant local tangent space alignment algorithm （KOTSDA）, is proposed. Firstly, the kernel mapping is performed to map the face data into a high dimensional nonlinear space and extract the nonlinear information. Then, tangent space discriminant analysis algorithm is used to preserve the intra-class local geometric structures and simultaneously maximize the inter-class difference in target function. Finally, KOTSDA is obtained with orthogonal constraints. It effectively avoids losing discriminant information which does not need to preprocess by PCA dimensional reduction. The experiments on ORL and Yale face databases demonstrate the effectiveness of the proposed algorithm.
出处
《模式识别与人工智能》
EI
CSCD
北大核心
2013年第7期673-679,共7页
Pattern Recognition and Artificial Intelligence
基金
国家自然科学基金资助项目(No.61272258)
关键词
特征提取
局部正切空间对齐
核空间
流形学习
Feature Extraction, Local Tangent Space Alignment, Kernel Space, Manifold Learning
