【摘要】 提出了一种基于核的全局正交鉴别矢量集方法。首先应用支持向量机中核函数的概念,将样本隐式地映射到特征空间,然后构造特征空间的正交空间,再将特征空间样本映射到特征空间的正交空间,求解基于Fisher的全局正交鉴别矢量集。在ORL人脸图像库上的实验结果表明,与现有的基于核的广义鉴别矢量集算法(GDA)和改进的基于核的广义鉴别矢量集算法(MGDA),以及基于核的直接鉴别矢量集算法(KDDA)相比,本文算法有更高的识别率及鲁棒性。更多还原

【Abstract】 A kernel-based method to find global orthogonal discriminant vectors was proposed.This method first maps training data into a high dimensional feature space through kernel function concept in Support Vector Machine and constructs orthogonal space in the feature space.Then it maps the data from feature space into the orthogonal space and computes Fisher criterion based global orthogonal discriminants.Experiments on ORL face image datasets show that compared with current general discriminant algorithm(GDA),modified general discriminant algorithm and kernel based direct discriminant algorithm,the proposed method has a better performance in precision and robustness.更多还原