摘要
针对人脸识别这一非线性分类问题,提出了一种基于核的无相关鉴别矢量集算法。应用了支持向量机中核函数的思想,通过核映射将原空间的非线性分类问题转化为特征空间的线性分类问题,然后在特征空间进行无相关鉴别矢量集的求取。其优势在于:利用核函数不但可以将非线性问题转化为线性问题,而且可以提取样本图像的高阶统计特征。在ORL人脸库中的测试结果表明,与传统的全局正交鉴别矢量集算法及传统的无相关鉴别矢量集算法相比,基于核映射的无相关鉴别矢量集算法有更高的识别率,最高识别率可达到99%。
An algorithm for the uncorrelated discriminant vectors set based on the kernel mapping was proposed to solve the nonlinear classification problem of the face recognition. Applying the concept of the kernel function in the supporting vector machine, the nonlinear classification problem in the original space was transformed to the linear classification problem in the feature space by the kernel mapping, and the uncorrelated discriminant vectors set was solved in the feature space. The advantage of the algorithm consists in that the nonlinear problem can be transformed to the linear one, and the high order statistical features among the pixels of the face image can be extracted at the same time. The experiments in the ORL face database showed that the proposed algorithm is characterized by a higher recognition rate than the traditional algorithms for the global orthogonal and the uncorrelated discriminant vectors sets, and the highest recognition rate may reach 99%.
出处
《吉林大学学报（工学版）》
EI
CAS
CSCD
北大核心
2006年第4期574-578,共5页
Journal of Jilin University:Engineering and Technology Edition
基金
国家自然科学基金资助项目(60372060)
科技部国际合作计划项目(2005DFA10300)
关键词
信息处理技术
人脸识别
Fisher无相关鉴别矢量集
核映射
information processing
face recognition
Fisher uncorrelated set of discriminant vectors
kernel mapping
